A Web-Interface Based Decision Support System for Optimizing Home Healthcare Waste Collection Vehicle Routing

Abstract

Background: The significant increase in home healthcare (HHC) driven by technological advancements, an ageing population, and heightened disease outbreaks—especially evident during the COVID-19 pandemic—has created an urgent need for improved medical waste management. Methods: This paper presents the development of a decision support system with a web-based interface designed for efficient medical waste collection in the HHC sector. Results: The system utilises Flask for backend operations, with HTML and CSS for the user interface, and manages data using JSON files. Its flexible design supports real-time adjustments for various vehicle types and changing waste production locations. It incorporates dynamic routing by employing two sophisticated metaheuristic algorithms: the Strength Pareto Evolutionary Algorithm (SPEA-2) and the Non-Dominated Sorting Genetic Algorithm (NSGA-II). This setup supports different dataset sizes and vehicle fleets, including Internal Combustion Engine (ICE) vehicles and Electric Vehicles (EVs). Conclusions: The automation reduces uncertainties in waste collection by minimising human intervention. The system is built to be easily adaptable for other sectors with minor modifications and can be expanded to test various scenarios with new selectable parameters.

1. Introduction

The Travelling Salesman Problem (TSP) is a critical optimisation challenge in combinatorial mathematics and operations research, focusing on finding the shortest or least costly route for a salesperson to visit each city in a list exactly once and return to the starting point [1]. Building on the framework of TSP, the Vehicle Routing Problem (VRP) introduces a more complex scenario with multiple vehicles and specific demands at each location. The aim is to devise optimal routing strategies for each vehicle to satisfy demands across all nodes, thereby minimising the cumulative operational costs [2]. Since Dantzig and Ramser [3] introduced the VRP in 1959, it has seen extensive application across various fields, initially concentrating on forward logistics and expanding to include reverse logistics operations. Since the introduction of the first Periodic Vehicle Routing Problem (PVRP) analysis for municipal waste collection by Beltrami and Bodin [4], the application of vehicle routing problems in reverse logistics has been extensively adopted. The waste collection process represents a key area within reverse logistics, focusing on optimising routing strategies [5]. Medical waste collection has become a critical research topic within waste management, gaining increased significance after the COVID-19 pandemic [6,7,8].

Recently, the generation points of medical waste have extended beyond hospitals and healthcare institutions to encompass the growing domain of home healthcare (HHC) services [5]. This development has introduced challenges in establishing robust coordination between decision-makers and system entities compounded by rising uncertainty and complexity. An early study in the HHC sector proposed a deposit-refund system for disposing of syringes, presenting an economically and environmentally viable approach to minimise the hazards of incorrect waste handling [9]. Another significant advancement in the literature on medical waste in home healthcare is the introduction of a system that monitors patients’ medical treatments and facilitates the safe disposal of medical waste at home through a mobile application [10]. However, to optimise this process and ensure comprehensive waste management, it is crucial to develop and implement a detailed vehicle routing plan for the efficient and systematic collection of these wastes [5]. For example, Ji [11] introduced the Simultaneous Delivery and Pick-up (VRPSDP) model to solve a vehicle routing challenge that involves distributing particular medications from hospitals to patients, delivering drugs from pharmacies to patients, and retrieving unused medical supplies from patients’ homes. The increasing complexity and diversity of waste collection in the HHC context demands innovative solutions that integrate technological advancements and real-time data to ensure efficiency and adaptability.

While the literature abounds with mathematical models addressing the waste collection routing problem, there is a distinct need for practical studies that offer sustainable and efficient routing plans capable of adapting to market uncertainties. Vitorino de Souza Melaré, Montenegro González [12] highlighted the critical need for a smart Decision Support System (DSS) in the waste management area, underscoring its importance for the practical aspects of research implementation. Santos, Coutinho-Rodrigues [13] argued that such a system is essential for improving the efficiency, sustainability, and effectiveness of waste collection routing processes. Within this context, Burton Watson and John Ryan [14] introduced a web application interface-based DSS designed to generate and showcase dynamic, resilient routes. This interface adapts to changes in the system by accessing updated data. The fluctuating nature of waste accumulation and disposal sites, which evolve rapidly over time, can be effectively captured and demonstrated through a practical study involving designing a web interface-based DSS. The autonomous and dynamic features of web-based DSSs enhance the management of waste collection routes by adjusting the number of vehicles required, driver assignments, and collection schedules [15]. In practical scenarios, factors like travel times impacted by traffic congestion, quantities of waste at generation points, and the locations of waste generation introduce significant fluctuation in resource planning within the context of DSS. Gasque and Munari [16] asserted that web interface applications serve as highly effective tools for aiding decision-making processes related to pickup and delivery tasks and can also be utilised as a framework for different versions of the vehicle routing problem. Traditional vehicle routing models cannot facilitate efficient and accurate decision-making processes due to their inherent limitations in responding to systemic changes on time.

This study aims to address these challenges by introducing a web interface-based Decision Support System (DSS) designed specifically for the HHC sector, capable of dynamically adapting to fluctuations in waste generation and service demands. The DSS leverages advanced routing algorithms and user-friendly interfaces to enhance decision-making processes, significantly contributing to both theoretical models and practical implementations in waste management logistics. Our approach represents a pioneering effort to integrate web-based technologies with vehicle routing problem frameworks to create a responsive and scalable system that meets the evolving needs of modern healthcare environments. Consequently, new web-based DSSs have been introduced in this study to accommodate changes in various factors, including travel time, the volume of waste to be collected, and waste generation points, while dynamically presenting routes to the drivers (users) in the HHC sector for the first time.

The remainder of this paper is structured to provide a comprehensive understanding of the challenges and solutions in modern waste management. Section 2 provides a detailed literature review, examining the web interface application-based DSS models in forward and reverse logistics. Section 3 presents brief information on the system description, mathematical model, and the detailed web interface-based DSS application process. The paper then progresses to Section 4, discussing the theoretical and managerial implications of the findings, thereby highlighting the contribution of this research to the field. The paper concludes with Section 5, summarising the key findings, implications, and potential directions for future research.

2. Literature Review

The application of the TSP to HHC logistics seeks to enhance route efficiency for healthcare providers [17,18,19,20]. Although TSP modelling sufficiently addresses fundamental healthcare visitation requirements, collecting healthcare-generated waste from residential settings necessitates the implementation of VRP modelling, an advanced derivative of TSP [21]. The increase in demand for home healthcare services has amplified the complexities of medical waste management, extending its relevance from medical facilities to residential environments and expanding the range of considerations within this field. As a result, a significant focus has been on developing the capabilities of VRP models tailored for medical waste collection [5].

Web interface application-based DSS tools for waste collection routing serve as a crucial link between theoretical models and their practical or, more specifically, commercial applications. These technologies improve the efficiency and effectiveness of waste management routing by transforming uncertain parameters, such as travel times affected by traffic congestion and variable waste accumulation rates, into deterministic factors through real-time data acquisition [22]. Furthermore, user-friendly route maps accessed through web pages facilitate decision-making and enhance management by enabling efficient planning [14]. On the other hand, Epifancev, Hvan [23] pointed out the scarcity of literature on waste management software, particularly noting the absence of features like environmental monitoring and routing planning in existing applications. Bányai, Tamás [24] and Sar and Ghadimi [5] outline the smart approach to waste collection routing, as illustrated in Figure 1. The diagram shows that travel times and waste bin fill levels are pulled from the database, activating the routing algorithm. This information is directly delivered to the driver (end-user) through a web interface.

The approach for smart vehicle routing showcased in Figure 1 has been applied and validated in numerous sectors, focusing specifically on forward logistics rather than reverse logistics processes. The comprehensive analysis of the literature review is presented in

Table 1. Summary of Previous Research Papers on Web Services based Vehicle Routing Problem.

Author	Problem	Solution Algorithm	Logistics Process	Sector	Interface
Nacakli, Guzel [25]	Two-Dimensional Vehicle Pallet Loading with Routing	MaxRects, Skyline and Guillotine with Dijkstra’s algorithm	Forward	Building and Construction	FLASK
Gasque and Munari [16]	VRP with pickup and delivery	Adaptive Large Neighbourhood Search	Forward and Reverse	E-commerce	JavaServer Faces
Burton Watson and John Ryan [14]	Smart Bin based VRP	-	Reverse	Solid Waste	FLASK
Tsoukas, Boumpa [26]	CVRP-TW	Modified Hopfield Network	Forward	E-commerce	FLASK
Moeini and Mees [27]	TSP	Route Generation Heuristic and Driver Assignment Heuristic	Forward	School Tour Planning	FLASK
Tagorda, Calata [28]	CVRP	Two-phase Heuristics	Forward	E-commerce	FLASK
Our Study	Smart CVRP *	NSGA-II and SPEA-2	Reverse	Medical Waste	FLASK

* Our study mathematical model can be found in [5].

.

For example, Nacakli, Guzel [25] developed a web-based interface DSS application for vehicle loading and routing problems in the forward logistics area to address the scarcity of practical research. The goal was to devise a time and performance-efficient method that surpasses conventional routing plans. Tsoukas, Boumpa [26] developed an e-commerce vehicle routing application, focusing on the forward logistics process. This web interface-based DSS integrates multiple algorithms to optimise routing, demonstrating flexibility in adapting to live traffic conditions. Mendes and Iori [29] developed and introduced a web-based DSS tool designed to optimise the routing process using a multi-trip VRP model specifically tailored for a pharmaceutical distribution company. These web interface application studies within forward logistics aim to enable sustainable and efficient planning by transforming theoretical models into practical tools for commercial use [26]. Research on waste management vehicle routing has remained largely theoretical due to the need for more commercial objectives. However, some projects have tackled practical elements through web interface DSS applications to enhance solid waste collection, supported by municipal partnerships. For example, Burton Watson and John Ryan [14] proposed a VRP model tailored for the solid waste collection problem in Melbourne, Australia was addressed using a heuristic solution approach. Also, a practical application was developed, providing real-time access to waste bin fill levels and adaptable routing plans via a web interface-based DSS. In Valdivia, Chile, a municipality-initiated project was designed as a web and mobile application leveraging Free and Open Source Software (FOSS) to tackle challenges in solid waste collection. The application aims to improve community engagement with waste services, real-time garbage truck tracking, illegal dumping reporting, and service status updates [30]. In Sri Lanka, a new practical study called Vehicle Routing Problem with Time Windows (VRPTW) was introduced to enhance solid waste collection, and a solution called Large Neighbourhood Search (LNS) was implemented for the solution phase. This model is integrated with developing a web interface based DSS that dynamically utilises real-time data to adjust routes based on the current waste production needs [31].

These practical studies, infrequently discussed in the existing literature and primarily tailored for solid waste collection, utilise basic mathematical models in their application. Nevertheless, it necessitates a more comprehensive integration of sustainability considerations and practical constraints. For this reason, this research introduces a smart multi-vehicle routing model that accounts for capacity, time windows, and specific waste thresholds, aiming to embed sustainability across social, environmental, and economic dimensions. Utilising the Internet of Things (IoT) for waste level monitoring, the model optimises route efficiency while adapting to real-time traffic data to reflect current conditions accurately. Different metaheuristics solution algorithms were developed to test various scenarios to obtain the most near-optimal solutions. Accordingly, a new DSS, built on a Python-based FLASK backend, have been proposed in this paper, which deploys the developed model. Such a DSS development in the home healthcare waste collection route planning research domain was not sufficiently explored in the previous research activities.

In conclusion, the primary contribution of this work is to develop a web interface-based DSS model approach for waste collection vehicle routing problems to provide better communication and structured information exchange processes which helps the drivers inside waste collection organisations to make better sustainable sourcing decisions in a more prompt and less human-interacting manner, resulting in maintaining a long-term partnership among drivers and collection points.

3. The Case Study

The case study presented in this paper concerns a medical waste collection routing issue encountered by an HHC service provider. It has been thoroughly examined, and subsequent sub-sections propose an in-depth system overview, an advanced mathematical model, and a DSS.

3.1. System Description

In the examined case study, an HHC service provider delivers medical supplies to its clients through an IoT-enabled box, which includes an integrated waste bin. In a previous research study by Sar and Ghadimi [32], a reverse logistics network was developed for this case study, where these waste bins can be retrieved either from centralised collection points, or directly from the residences of customers who lack access to these collection facilities. In the collection routing phase, an extensive mathematical model was developed that addresses all aspects of sustainability, incorporating various real-world constraints including time windows, capacity limits, maximum working hours, the possibility of multiple rounds, a variety of vehicle types (such as Internal Combustion Engine vehicles and EVs), and differing rates of waste accumulation. The foundational phase of the described model can be found in Sar and Ghadimi [5]. Accordingly, Section 3.2 presents the advanced version of the developed model.

3.2. Mathematical Model

This section outlines the proposed mathematical model mentioned in the previous section. The notation of the mathematical model can be found in

Table 2. Notations for Vehicle Routing Problem Mathematical Model.

Sets	Description
K	Set of all vehicle types, K = {1, 2,… k,…}
T	Set of trips/routes T = {1, 2,… t,...}
N	Set of all collection points N = {1, 2, 3…n, n + 1} depot = {0}
Decision Variables	Description
$x_{ijk}^t$	Binary variable with a value of 1 if arc (i,j) is traversed of vehicle k on trip t.
$T_{jk}$	Actual arrival time at node j by vehicle k
$Q_{ijk}$	Carried load (kg) of vehicle k visit from point i to point j
Parameters	Description
$d_{ij}$	Distance between nodes i and j.
$t_{ij}$	The travel time between nodes i and j.
$c_p$	Penalty cost per unit of time
${LT}_j$	Latest acceptable arrival time at node j
${\eta }_0$	Amount of fuel consumed per km when the ICE vehicle is fully empty (kg/L)
η	Amount of fuel consumed per km when the ICE vehicle is fully loaded (kg/L)
$P_{idle}$	Amount of fuel consumed per minute when the ICE vehicle is running idle
$s_i$	Service time at collection points
$e$	Emission coefficient
Q	Maximum cargo capacity of the vehicle

.

3.2.1. Objective Function 1: Minimisation Travel Time

In the literature, the goal of minimising total travel distance in VRP is often associated with economic benefits. However, this approach may not always be the most efficient one, particularly in situations with traffic congestion. Additionally, the impact of EVs on total travel time is a significant aspect that warrants further investigation. Therefore, the multi-vehicle multi-trips smart capacitated vehicle routing model with a time window and threshold waste issue has been proposed by prioritising the reduction in total travel time (as depicted in Equation (1)), with a particular focus on economic outcomes. The Google Distance Matrix API was used to ensure precise calculation of travel times, incorporating traffic conditions into consideration.

$$min Z_1={\sum }_{t=1}^T{\sum }_{k=1}^K{\sum }_{i=0}^N{\sum }_{j=0}^N{x}_{ijk}^t . t_{ij}$$

(1)

3.2.2. Objective Function 2: Minimisation of Total Emission

Carbon emissions play a significant role in logistics concerns, mainly when using vehicles with ICE [33]. The adoption of EVs presents a way to address these emissions, though it comes with challenges. Consequently, the second objective function of the model is aimed at reducing total carbon emissions, highlighting the environmental significance of using both ICE and EVs. Carbon emissions correlate directly with the distance a vehicle travels, as emissions per distance are fixed but vary according to the type of vehicle and its fuel. Therefore, greater distances lead to increased emissions.

Furthermore, emissions generated during the loading and unloading process at collection points, which are frequently overlooked, are accounted for in this analysis. This consideration encompasses the emissions produced by vehicles while idling at collection centres. The sum of driving and idling emissions gives us the total emissions as in Equation (2), steering towards more eco-friendly and sustainable logistics operations.

$$min Z_2=C_2$$

(2)

The total carbon emissions are calculated using the proposed method by Xiao, Zhao [34], as detailed in Equations (2a)–(2c). Equation (2a) calculates fuel-driven emissions, factoring in the load and distance travelled. Equation (2b) accounts for idling emissions at each collection point. Equation (2c) sums the driving and idling emissions to provide a total emission impact for each route.

$$C_{fuel-drive}={\sum }_{t=1}^T{\sum }_{k=1}^K{\sum }_{i=0}^N{\sum }_{j=0}^N({\mathrm{\eta }}_0+{\displaystyle \frac{\mathrm{\eta }-{\mathrm{\eta }}_0}{Q}}.Q_{ijk}).d_{ij}.x_{ijk}^t.e$$

(2a)

$$C_{fuel-idle=}{\sum }_{i=1}^N{x}_{ijk}^t{.s}_i{.P}_{idle}.e$$

(2b)

$$C_2{=C}_{fuel-drive}+C_{fuel-idle}$$

(2c)

Equations (2a) and (2b) calculate the total vehicle carbon emissions during waste collection, factoring in driving and idling. Equation (2a) accounts for the vehicle’s fuel efficiency for driving emissions, which varies with load and the distances travelled between collection points. Idling emissions, Equation (2b), are determined by the time spent idling at collection sites, with a specific fuel consumption rate for idling. The total emissions (Equation (2c)) are the sum of these two factors, and the objective is to minimise this total (Equation (2)), aiming for more environmentally friendly logistics operations.

3.2.3. Objective Function 3: Minimisation of Social Risk

Previous studies have highlighted the economic efficiency of excluding low-waste sites from routing plans, showing that visiting locations with waste levels at or above 70% is more beneficial [35,36]. Our research expands on this by considering both the economic advantages and the social implications, which is mainly customer dissatisfaction due to excessive waste accumulation. A key strategy is prioritising high-waste areas over those with less waste. This is achieved through implementing the soft time windows in the routing plan, focusing first on areas with waste levels above 90% within the initial hour of the route. Non-compliance with these time windows results in penalty scores, reflecting social risks.

Furthermore, the extended recharging times for EVs could delay visits to areas with high waste levels, potentially leading to customer dissatisfaction. Penalty scores are applied for visits beyond the latest permissible time for each point, while there is no restriction on the earliest visit time, allowing points to be serviced from minute zero. In essence, minimising social risk is established as the third objective function, shown in Equation (3).

$$\min Z_3=S_{penalty}$$

(3)

The detailed equation description is shown in Equation (3a). The penalty score is defined as follows:

$$S_{penalty}=c_p.{\sum }_{t=1}^T{\sum }_{k=1}^K{\sum }_{i,j=1}^N{x}_{ijk}^t.max\{T_{jk}-{LT}_j),0\}$$

(3a)

3.3. Solution Algorithm

In this study, we employed the NSGA-II and the SPEA-2 to address the complex VRP optimisation challenges in home healthcare waste collection. The inherent complexity and NP-hard nature of VRP, mainly when they involve multiple objectives, such as minimising carbon emissions and addressing social issues alongside traditional logistics goals, necessitate robust and adaptable optimisation strategies. These problems are characterised by their computational intensity and the vastness of their search spaces, making identifying optimal solutions through conventional optimisation techniques or exhaustive searches impractical. In the realm of solving such intricate VRPs, a variety of metaheuristic algorithms have been developed and employed. These algorithms, including GA, PSO, and SA, are designed to navigate the complex solution landscapes of NP-hard problems efficiently. Each approach brings distinct advantages; for example, GA is renowned for its effectiveness in exploring extensive search spaces, PSO is valued for its quick convergence properties, and SA is adept at avoiding local optima, enhancing solution space exploration [37,38,39]. Among these, the Non-dominated Sorting Genetic Algorithm II (NSGA-II) is particularly noteworthy due to its capacity to handle multi-objective optimisation challenges inherent in VRP [40,41]. NSGA-II balances exploration and exploitation within the solution space, allowing for identifying a broad set of optimal solutions representing the Pareto front [42]. This capability is vital for VRP applications where trade-offs between various objectives—such as reducing travel time, minimising carbon emission, and mitigating social risk—must be optimised simultaneously [43].

Empirical studies, including those by Wang, Peng [44] and Aliahmadi, Barzinpour [45], have highlighted the superiority of NSGA-II in addressing multi-objective VRPs, demonstrating its effectiveness over alternative metaheuristic approaches in terms of solution quality and computational efficiency. These studies underscore the algorithm’s adeptness at managing large datasets, a common challenge in waste management VRP. Therefore, based on its demonstrated success and specific advantages for multi-objective optimisation, NSGA-II was chosen as the preferred method for addressing the model presented in this research. Its widespread application and proven effectiveness in similar contexts underscore its suitability for tackling the multifaceted challenges posed by VRP, particularly those aimed at optimising vehicle routes for waste collection with the dual goals of minimising economic concern and environmental issues. Therefore, NSGA-II was the initial algorithm used to solve this model due to its effective performance of multi-objective VRP models in the literature. The NSGA-II solution approach, developed by Deb, Pratap [46], has been deployed considering the following steps:

• Initialisation: The initial population consists of random solutions (chromosomes).
• The chromosomes (i.e., solutions) are evaluated. After each chromosome is assessed, it is ranked based on its performance. The best solutions are placed in rank 1 (F1), the second-best solutions are placed in rank 2 (F2), and so on. Therefore, rank 1 consists of the top-performing solutions.
• The crowding distance operator assesses the chromosomes with the same rank and calculates the average distance between neighbouring solutions. When two solutions have the same rank, the one with a greater crowding distance is selected. This process ensures that the Pareto solutions are evenly distributed. After completing this procedure, the parents that will be used to generate the offspring are identified.
• After the parent selection process, new generations are generated with mutation and crossover operators. These operators aim to increase the diversity of the new offspring.
• A new population is created by merging the newly generated offspring with the selected parents from the previous generation. These individuals are then sorted using non-dominated sorting criteria, and the best solutions are chosen based on their crowding distance and rank. Figure 2 shows how this procedure is implemented.
• The solutions assigned to rank one are saved as potential Pareto solutions.

This procedure is repeated until the maximum number of iterations is achieved.

-

Solution Representation

Figure 3 illustrates a random solution representation using a genetic algorithm for a vehicle routing problem with ten nodes and three vehicles. The figure depicts the initial configuration of chromosomes, where each row corresponds to a different vehicle’s route. The first element in each row indicates the vehicle identifier, which starts from the depot. The subsequent numbers represent the sequence of customer nodes assigned to that vehicle. This configuration highlights how each vehicle’s route is determined from the depot, serving assigned customer nodes until the vehicle’s capacity or the route’s maximum duration is reached. The allocation sequence respects the vehicle’s capacity constraints, and when a vehicle reaches its limit, the next vehicle/tour takes over in the sequence. This genetic representation is crucial for understanding the solution’s initialisation phase, where random solutions are generated before the optimisation process begins. The genetic approach helps to explore diverse possible solutions, enhancing the algorithm’s ability to find more efficient routes under complex constraints such as capacity limits and time windows. The time window constraint, treated as a soft constraint, does not strictly limit the routes but incurs penalties for early or late arrivals, which are factored into the fitness evaluation of each solution.

-

Crossover and Mutation Operators

Several mutation and crossover operators are used in the GAs and are selected depending on the problem. The Order Crossover (OX) approach was applied in this study. The crossover operator applies after choosing the parents, with chromosomes with better ranks and crowding distances more likely to be selected. The OX method randomly selects two genes from the parents’ chromosomes once the parents have been determined. Then, offspring 1 directly receives the part between these two genes in the first parent. Offspring 1’s infilling genes are completed by utilising the corresponding genes from the second parent. This exact procedure is carried out for offspring 2 and parent 2. The operational mechanism of the OX in sample chromosomes is illustrated in Figure 4.

The crossover operation produced offspring 1 and offspring 2, who were formed from parent 1 and parent 2. Afterwards, the mutation operator was used to increase the diversity. There are different types of mutation operators. The swap mutation operator was applied in this study. In the offspring’s chromosomes, the swap operator randomly chooses two genes and switches their locations. It is a straightforward but efficient approach to add variation to the population, which can help keep the algorithm from being trapped in local optimums. The operational concept of the swap mutation is demonstrated in Figure 5.

The SPEA was developed by Zitzler and Thiele [47]. It was one of the first evolutionary algorithms specifically designed for multi-objective optimisation problems. The algorithm gained significant attention due to its innovative fitness assignment strategy based on Pareto dominance and its external archive mechanism for preserving non-dominated solutions. The SPEA and its advanced version, SPEA-2 which was developed by Zitzler, Laumanns [48] have proven to be highly effective in addressing multi-objective optimisation challenges within VRP mathematical models. These algorithms are noted for their superior capability to handle complex, competing objectives, leading to more efficient and optimal solutions in VRP scenarios. For example, Rabbani, Nikoubin [49] solved a multi-objective VRP model using SPEA-2 and found that it provided strong results in terms of optimising cost, travel time, and service quality. By employing comparison metrics, they demonstrated that SPEA-2 outperformed other approaches in effectively balancing these conflicting criteria within the VRP framework. In the scholarly landscape, SPEA-2 has been extensively employed across numerous studies focusing on both forward and reverse logistics VRP, showcasing its robust capability in managing multi-objective optimisation challenges inherent in vehicle routing problems. This algorithm distinguishes itself by adeptly harmonising competing objectives. This capacity of SPEA-2 to mediate between conflicting objectives and produce superior outcomes makes it a valuable tool for advancing research in the domain of logistics optimisation [50,51,52]. The methodology employed in SPEA-2 closely mirrors that of NSGA-II and can be systematically described as follows [48]:

• Initialisation: Generate an initial population of solutions randomly.
• Fitness Assignment: Evaluate the fitness of each individual in the population. In SPEA-2, this involves both the individual’s objective values and a measure of how many solutions it dominates or is dominated by.
• Environmental Selection: Select the best individuals to form a new population, using a combination of Pareto dominance and density estimation to maintain diversity.
• Binary Tournament Selection: Use binary tournament selection based on fitness to select parents for reproduction.
• Crossover and Mutation: Apply genetic operators such as crossover and mutation to generate new offspring.
• Update External Archive: Update an external archive that stores the best non-dominated solutions found during the search process.
• Termination: Repeat the process until a termination criterion is met, which could be a set number of generations or a convergence threshold.

Overall it can be said that both SPEA-2 and NSGA-II are commonly used multi-objective solution methods in the literature for VRP problems, and their performances are often compared [20,50,52,53]. Thus, this section compares the NSGA-II solution method preferred in this model with the SPEA-2 algorithm in terms of various factors, such as computational time, number of discovered solutions, spacing, and diversity. In order to obtain results in a reasonable time in testing the performance of algorithms, both the SPEA-2 and NSGA-II algorithms were configured with a population size of 20 and a generation number of 10.

(1) Number of found solutions (NOS): The number of solutions found in the Pareto front are essential factors in multi-objective optimisation problem solutions as they provide insight into the quality of the solution. Algorithms that can generate a greater number of solutions, meaning enabling diversity, are high performing. That is why, the number of solutions generated by SPEA-2 and NSGA-II was compared and evaluated. According to the data in

Table 3. Number of found solutions for each algorithm.

	Small Dataset	Medium Dataset	Large Dataset
NSGA-II	14	13	7
SPEA-2	55	58	24

, SPEA-2 outperformed NSGA-II significantly in terms of the number of found solutions for both small, medium, and large datasets. Specifically, it can be said that SPEA-2 generated roughly four times as many solutions as NSGA-II.

(2) Spacing: The spacing metric is a commonly used measure to assess how evenly the points on the achieved Pareto frontier are distributed [49]. A smaller value of the spacing metric is preferred as it indicates a more evenly distributed Pareto frontier. As in

Table 4. Results of spacing metric performance measures.

	Small Dataset	Medium Dataset	Large Dataset
NSGA-II	13.79	9.01	68.97
SPEA-2	6.005	9.299	255.58

, for the medium dataset, there is not a significant distinction in performance between the algorithms, and they can be evaluated equally. For the small dataset, SPEA-2 outperforms NSGA-II. Conversely, in the large dataset, NSGA-II demonstrates superior performance compared to SPEA-2.

(3) Diversity: A diversity metric is used to quantify the extent of the Pareto front, which represents the distribution of non-dominated solutions. In general, a larger value of the diversity metric is considered desirable as it indicates a greater spread of the Pareto front [54]. According to the results shown in

Table 5. Results of diversity metric performance measures.

	Small Dataset	Medium Dataset	Large Dataset
NSGA-II	0	0	790
SPEA-2	0	0	1100

, neither algorithm achieves the expected performance standards for this metric in the small and medium datasets. However, for the large dataset, SPEA-2 shows a slight improvement over NSGA-II.

(4) Computational time: In the realm of optimisation problems, minimising computational time is often considered a crucial metric for evaluating the performance of proposed or existing algorithms. As such, the ability to achieve quicker results is often deemed favourable and can provide valuable insights into the efficiency and effectiveness of an algorithm in addressing a particular problem. The performance of the NSGA-II and SPEA-2 algorithms was evaluated and compared in terms of computational time for our proposed VRP model.

Table 6. Computation time for each algorithm.

	Small Dataset	Medium Dataset	Large Dataset
NSGA-II	0.77	0.53	1.31
SPEA-2	663	759	1141

presents the total computational time in seconds for each algorithm across all datasets. The NSGA-II algorithm outperformed the SPEA-2 algorithm by a large margin. Specifically, for small and large datasets, NSGA-II was almost 900 times faster than SPEA-2, while for medium datasets, NSGA-II was nearly 1500 times faster than SPEA-2.

3.4. Decision Support System Development

The proposed DSS has metaheuristic solution algorithms to optimise the user’s (decision-maker) routing paths. Currently, the metaheuristics algorithm choices, which are NSGA-II and SPEA-2, are available through a selectable option. This system is designed with future scalability in mind, allowing for the integration of additional algorithms to enhance solution capabilities as needed. This DSS offers the flexibility of catering to different environmental contexts by allowing users to select from conventional ICE vehicles and EVs. The DSS is designed to handle data that changes in real-time using JavaScript Object Notation (JSON) files. To ensure it works effectively across various scales, it is essential to conduct thorough testing with datasets categorised as small, medium, and large, utilising the sub-interface for data size selection. Looking ahead, the interface will maintain a feature for selecting data for activities that cover larger geographical areas, allowing users to obtain the optimum route plan for their operational region. This proposed interface provides a streamlined user experience for configuring and running VRP simulations, visualising results, and examining the potential environmental and economic impacts of different vehicle routing strategies.

3.5. Core Functionalities

The web interface-based DSS platform, built on Flask, provided a solid backend structure, offering an effective method for solving the VRP using the chosen metaheuristic algorithm. Flask is a compact, micro web application framework developed in Python. It offers fundamental tools for developing web applications and services, allowing developers to choose their application’s structure without prescribing specific components [55]. The primary route, known as the root route, efficiently handles both GET and POST requests. While the GET request displays the index.html page, the POST request delves deeper. It receives three crucial files: input data detailing customer location, waste metrics, and time windows, along with a mathematical model file and execution files.

The core functionality of the proposed system is encapsulated within two primary Flask routes. The first route, accessed via a ‘GET’ request to the root URL (‘/’), serves the initial configuration form to the user. This form, defined in ‘index.html’, allows users to specify the type of vehicles (ICE or EVs), the size of the dataset for the VRP (Small, Medium, or Large), and the solution algorithm (NSGA-II or SPEA-2). This makes the complex VRP model a practical tool accessible to all users, enabling them to leverage its capabilities without facing technical challenges. Upon submitting the form, a ‘POST’ request is sent to ‘/vehicle-routing-result’, triggering the execution of the VRP solver based on the selected parameters. The system dynamically loads the appropriate dataset and invokes the ‘solution class’, which implements the NSGA-II or SPEA-2 metaheuristics algorithm tailored for solving VRPs. These algorithms efficiently find near-optimal routes that balance multiple objectives, such as minimising total travel time, emissions, and customer dissatisfaction. The results, including detailed route information and performance metrics, are then rendered in ’vehicle_routing_result.html’ using a responsive design that adapts to various display sizes. These results page visually represents the VRP solution, incorporating an embedded map (via an iframe) displaying the optimised routes.

Integrating Google Maps and Folium for route visualisation equips users with intuitive, interactive maps. These maps display optimised routing paths with markers for various points of interest and colour-coded routes to distinguish between different vehicle types. This not only aids in understanding the algorithm’s output but also in assessing the practical implications of the proposed routes in real-world settings.

In conclusion, the proposed DSS and underlying metaheuristics solution algorithm-based VRP solver significantly advance waste collection routing technologies with a special focus on HHC services. The user-friendly interface simplifies complex optimisation tasks, making robust routing optimisation accessible to a broader audience. This approach makes advanced computational tools more accessible, promoting sustainable and efficient logistics planning.

3.6. Design Architecture and Components

In the design of our web application, a clear and organised three-layer architecture, as demonstrated in Figure 6, was adopted. The system architecture commences with the “Interface Layer”, which can be equated to a corporate front desk, providing the initial interaction point for user input and preliminary results display. The design of this layer emphasises ease of use and clear presentation of information. Following this, the “Technical Layer” is instituted, operating as the engine room where the intricate data processing is executed. Embedded within are the solution algorithm and the mathematical model, both pivotal in transforming user inputs into actionable outputs. The final structural component is the “Data Resources Layer”, which acts as a depository, safeguarding the input data. This ensures that data are preserved and readily available for future analysis and application. Together, these layers form a coherent system engineered to facilitate each segment’s role in achieving the targeted result, as elaborated in the comprehensive exposition of the web interface development.

The web interface platform consists of two components: the front end and the back end [26]. From a front-end perspective, elements were showcased using Hypertext Markup Language (HTML). The visual design was articulated through Cascading Style Sheets (CSS), while critical scripting operations were conducted via JavaScript. Figure 7 illustrates the frontend component of the system.

The system includes a dedicated “/vehicle-routing-result” route, designed to display the optimal routes, visual mapping, and objective metrics based on the user inputs and chosen solution strategy. The initial part of this route provides essential metadata and links to resources like CSS and JavaScript, with a particular emphasis on the Leaflet library for map-based results. Additionally, the optimal sub-routes for individuals are outlined for clear understanding. A key feature of the results page is its map-based visualisation. A dedicated section, enhanced by JavaScript integration, uses Leaflet to depict the best sub-route distinctly. Each sub-route is colour-coded, with markers for precise coordinates, offering a clear and integrated display. The map automatically adjusts to encapsulate all the sub-routes, marrying the ease of web interface navigation with advanced algorithmic output, representing a step forward in VRP solutions through metaheuristic techniques. Figure 8 displays a sample of the routing outcome generated by the developed interface.

3.7. Validation and Key Performance Indicators (KPIs)

In the validation process of the web interface-based DSS designed for optimising home healthcare waste collection, thorough testing was conducted to ensure the system’s functionality and performance adapted to various operational scenarios. This involved the integration of real-time data inputs, such as dynamic waste generation rates and fluctuating traffic conditions, to validate the system’s responsiveness and accuracy. Through the use of sample data, simulations were carried out to refine the algorithms and validate their efficacy against established benchmarks and actual outcomes, allowing for necessary adjustments and optimisation. Key Performance Indicators (KPIs) played a crucial role in quantitatively assessing the system’s effectiveness and efficiency across multiple dimensions of the vehicle routing process, focusing on enhancing sustainability and optimising operations. These KPIs included “Total Travel Time Reduction” which measures the decrease in travel time compared to traditional routing methods, thus highlighting improvements in operational efficiency. “Emission Levels” were monitored to assess the environmental impact, comparing carbon emissions from Internal Combustion Engine (ICE) vehicles and Electric Vehicles (EVs) to underline the system’s contribution to environmental sustainability. “Service Level Accuracy” was another critical indicator, measuring the system’s reliability in adhering to predetermined customer time windows, thereby ensuring high service quality and customer satisfaction. Additionally, “System Scalability” was evaluated to determine the system’s capability to handle varying dataset sizes and adapt to changes in operational scope, which is vital for assessing its applicability to different or expanding service areas. The continuous monitoring of these KPIs during the testing phase provided critical insights into the system’s performance, identifying areas for improvement and confirming that the DSS not only meets but also surpasses the operational demands of home healthcare waste collection. This validation process underscores the potential of the DSS to transform logistics in the HHC sector, promoting a more sustainable, efficient, and user-friendly approach to waste collection.

4. Discussions, Theoretical and Managerial Implications

4.1. Discussions

This section evaluates the model’s performance and its broader implications for medical waste management vehicle routing. We explore the operational benefits of the proposed DSS, particularly its success in enhancing route efficiency and reducing costs. Despite these improvements, we acknowledge critical limitations within our current framework. A primary limitation is the model’s use of travel time data from the Google Distance Matrix, which provides real-time travel times based on current traffic conditions. Although practical, our periodic application of these data during the routing period may not capture all minor or major changes in traffic conditions. Consequently, the model might not fully reflect the dynamic and sometimes unpredictable traffic fluctuations, potentially affecting the routing plans’ accuracy and efficiency. Moreover, Google Distance Matrix utilises historical data, which helps analyse travel time by considering factors like weather conditions, accidents, or other unusual situations, but it still may not be completely reliable for real-time adjustments without continual updates.

Additionally, the model’s reliance on predefined vehicle capacities may not effectively represent the variable load sizes and types typically encountered in waste collection. The adaptability of the model to sudden shifts in environmental policies or economic conditions is also a concern. We discuss the potential need for parameter adjustments or structural changes to enhance the model’s responsiveness to such fluctuations. By highlighting these limitations, we aim to provide a balanced perspective that not only underscores the strengths of our approach but also pinpoints essential areas for future research and development. This analysis fosters continuous improvement and refinement of the model to better meet community needs and adapt to changing circumstances. Furthermore, if we gain access to real case study company data, we can employ machine learning techniques to analyse waste generation patterns more accurately. This data-driven approach would allow us to optimise routing plans further and prevent overflow issues by predicting when and where waste is most likely to accumulate, leading to more precise service scheduling. We also consider potential model enhancements, such as integrating real-time data feeds to dynamically adjust routes based on current traffic and waste generation patterns. Exploring multi-modal transportation options can further optimise logistics and reduce environmental impacts.

These future directions demonstrate our commitment to evolving the DSS to address emerging challenges and broaden its applicability across various sectors, thus advancing the field of sustainable waste management.

4.2. Theoretical Implications

This study introduces an innovative method for addressing the waste collection routing problem by integrating live data through a proposed DSS, including waste collection frequencies, traffic updates, and customer positions. Web interface application-based DSS studies have predominantly been implemented in the forward logistics domain. By leveraging web interfaces, these DSS applications enhance decision-making capabilities through real-time data visualisation, user-friendly interaction, and the ability to adapt to dynamic logistical scenarios [25,26,29]. A few studies have developed web interface-based DSS tools for waste management, but these have been limited to solid waste [14]. However, these studies were primarily designed using basic Python libraries instead of tools like Flask, HTML, and CSS, which are essential for web application development. This limitation restricts the functionality of the studies and significantly hinders their potential for expansion. Additionally, recent studies have only incorporated basic models into their DSS tools, lacking the integration of complex mathematical models and advanced solution algorithms for multi-objective optimisation in waste routing [56]. On the other hand, this developed interface helps establish robust coordination between those making decisions and the system parameters. Also, it enhances efficiency by minimising manual intervention, saving both time and money, while also reducing the likelihood of human errors. The results are presented with easy-to-understand map visuals and the ultimate figures of the objective function, highlighting a shift towards more automated processes. While the system is built to operate in real-time based on user inputs, its efficiency was evaluated using three distinct-size datasets. Across all three datasets, the outcome, including the routing map and objective function details, was generated, and displayed within a timeframe of under 30 s. While this web-based DSS application is tailored to the waste collection needs of the HHC sector, its adaptable design allows it to cater to the waste collection challenges of various other sectors with slight adjustments.

The developed web interface based DSS features one of the most comprehensive mathematical models developed for waste collection routing. This model incorporates a range of constraints and parameters such as waste threshold levels, real-time travel duration, integration of all sustainability concerns, and multiple vehicle and trip options, offering solutions for both EVs and ICE vehicles. The advancement of this web interface elevates its theoretical impact by offering an array of selectable parameters and leveraging cutting-edge solution algorithms, all built upon one of the most detailed mathematical models for waste collection routing. This interface streamlines the decision-making process, making it quicker and devoid of errors with minimal human input. Moreover, it offers a cost–benefit, enhances the quality of service, and improves the ability to adapt to changing variables.

4.3. Managerial Implications

The introduction of web-interface-based DSS tools in this study significantly enhances managerial decision-making in the home healthcare sector by providing a platform that integrates real-time data sources such as traffic updates, waste generation rates, and customer locations. This system empowers managers and users to make informed, dynamic decisions regarding waste collection routing, enabling them to adjust routes based on current conditions to optimise efficiency and reduce operational costs. For the case company, the integration of this system into their operations can be facilitated through IoT technologies for live data capture, ensuring that vehicle routes are continuously updated to reflect real-time scenarios. The technical requirements for successful integration include robust backend support, typically provided by a Flask-based framework, to handle the computational demands of real-time data processing and interface management. Additionally, the system’s flexibility to accommodate different vehicle types and dataset sizes ensures that it can adapt to various operational scales and needs. By reducing manual intervention and leveraging automated, algorithm-driven processes, the DSS not only improves operational efficiency but also enhances the overall sustainability and responsiveness of waste management practises in the home healthcare environment. This results in more timely waste collection, reduced environmental impact, and better resource utilisation, ultimately leading to improved service quality and customer satisfaction.

5. Conclusions, Limitations, and Future Works

This study has successfully developed a web interface-based DSS tailored to optimise medical waste collection within the HHC sector. This technology-driven approach significantly improves decision-making efficiency by minimising human intervention and enhancing automated processes. The system facilitates seamless transitions between different outcomes generated by various solution algorithms and allows for quick adaptation between EVs and internal ICE vehicles based on environmental priorities or fleet compositions. Despite these advancements, the study acknowledges certain limitations that must be addressed in future work. The DSS relies heavily on predefined algorithms and data inputs from the Google Distance Matrix for travel times and sensors in waste bins for real-time waste volume measurements. While these sources provide current data, the frequency of data updates and the inherent delay in processing and responding to this information may not fully capture the dynamic and often unpredictable nature of traffic conditions and waste generation. For instance, the Google Distance Matrix provides travel times that reflect current traffic conditions at the time of query. However, these conditions can change rapidly within minutes, affecting the accuracy and applicability of the routing solutions. Similarly, the data from waste bin sensors, while timely, represents waste levels only at the moment of measurement and may not account for significant, sudden increases in waste volume that occur between data transmissions. Additionally, the current model’s capacity to integrate and process these real-time data streams is limited, which could affect the accuracy and responsiveness of the routing solutions provided. The system’s ability to dynamically adapt to sudden changes in traffic and waste accumulation is therefore constrained, potentially leading to inefficiencies in route optimisation and scheduling.

Future directions for this research include incorporating additional solution algorithms and exploring new scenarios such as mixed vehicle fleets, to thoroughly evaluate various algorithms across different dataset scenarios. The integration of predictive analytics into the interface is another potential avenue, aiming to forecast waste generation patterns. This enhancement would facilitate proactive route optimisation and resource allocation, potentially leading to increased efficiencies and further environmental benefits. Moreover, augmenting current metaheuristic solution algorithms or incorporating innovative algorithms into the interface could provide additional options for optimisation. The system’s adaptable architecture allows for the inclusion of diverse or hybrid vehicle fleets and can be easily tailored to meet various waste collection challenges with minor adjustments, ensuring that it remains flexible and responsive to the evolving needs of waste management.

In conclusion, while the web interface-based DSS introduced in this study marks a significant advancement in leveraging modern technology for waste management, it also highlights the need for ongoing improvements. Addressing the system’s limitations and expanding its capabilities will ensure that it continues to meet the demands of this critical sector effectively.