Machine Learning, Data Mining, and IoT Applications in Smart and Sustainable Networks

The smart and sustainable networks require highly connected systems that can improve their operational performance, reduce environmental impact, and increase functional efficiency. The application of enabling technologies such as machine learning (ML), data mining, and Internet of Things (IoT) are of much interest to transform the traditional systems into smart and sustainable networks (e.g., smart farms, smart factories, smart cities, smart grids, and smart health, etc.) due to their huge potential [1,2]. However, application of IoT, data mining, and ML into smart and sustainable networks faces numerous challenges. Out of those, data security and privacy are the leading ones because such applications require extensive collection and analysis of data while the enabling IoT devices often have inadequate security that leaving them vulnerable to the cyberattacks [3,4]. Moreover, managing these large volumes of data is challenging that essentially requires efficient processing and storage solutions while maintaining data quality and integrity for accurate analysis. In this regards, steep data requirements and scalability concerns present significant hurdles in the development of smart and sustainable networks. In addition, integrating these new technologies into existing systems and ensuring IoT devices from different manufacturers work together adds complexity.

Despite all that, we need IoT devices that are low-power and cost-effective to combat sustainable design goals. Towards this end, energy consumption must be optimized to extend battery life while ensuring device functionality and environment protection [5,6]. Worse still, the initial costs of implementing these technologies, including sensors, actuators, storage and analytics, are high. On a technical level, IoT faces interoperability challenges, where seamless communication between devices from different manufacturers requires sophisticated solutions [7]. In IoT ecosystem, edge computing also requires efficient algorithms and powerful hardware architecture to process and analyze data at the edge of the network to overcome computing and storage limitations [8].

In data mining, ensuring data quality (handling missing values, outliers, and noisy data) is crucial and has a significant impact on the analysis results [9]. The complexity of data models and processing algorithms makes interpretation difficult and hinders the decision-making process. Similarly, ML presents unique technical challenges, including balancing model complexity to reduce overfitting or underfitting [10]. Interpretability of models, especially complex models such as deep learning, is critical in promoting trust and transparency. To prevent discriminatory results, it is crucial to eliminate bias and ensure the fairness of ML models. In this regards, hyperparameter tuning requires extensive experimental and computational resources to effectively optimize model performance. On the other hand, legal and regulatory hurdles arise from the need to comply with data protection laws, which vary from country to country. Last but not least, gaining user acceptance and trust requires careful management, as concerns about job transfer or privacy may hinder adoption of the technology. Therefore, experts in ML, data mining, and IoT technologies can provide advice, new findings, and advanced approaches to solving the problems of transforming conventional systems into smart and sustainable networks. In this context, we are constantly on the lookout for state-of-the-art high-quality papers based on ML, data mining, and IoT applications. We applied a rigorous peer-review process and selected published work from 50 submitted papers, which we briefly summarize here.

In this Special Issue, the use of ML methods for smart and sustainable systems is addressed by several articles. In Contribution 1, the authors demonstrate ML models using explainable artificial intelligence (XAI) for the detection of malicious domains. Therein, 45,000 samples were used with various interpretable ML and black box ensemble models. The authors claim 98% accuracy in detecting benign and malicious domains. In Contribution 2, the authors used ML-based feature selection to improve the electricity load forecasting of the smart grid. The assembly of an integration model along with a binary genetic algorithm is proposed to optimize the mean absolute percentage error. The method is potentially helpful in improving accuracy in the load scheduling of decentralized, small-scale generation units.

In Contribution 3, the authors present a review of the use ML algorithms for supervisory control and data acquisition in intrusion detection and classification systems. In this article, a compression of various supervised learning methods regarding datasets, feature engineering, methodologies, and optimization mechanisms and classification procedures is considered, which can help in the design of purpose-built security systems. In Contributions 4 and 5, we can find the uses of ML with some interesting applications related to the smart education and energy sectors.

In Contribution 6, the authors proposed an efficient multimedia content delivery architecture for a better quality of experience (QoE) and smooth operation of the Internet protocol. The authors claim that the proposed design improves the QoE of 4G/LTE users for multimedia data such as distribution, streaming, and banking operations. In Contribution 7, the author analyzed the relationship between the sustainable use and service quality of AI-based voice search technology through a user survey in Korea. The author found that the voice search service has a positive effect on the user’s interactivity in terms of sustainable use, playfulness, certainty, and empathy. This study has the potential to advance the technology used in the design of voice search services, where direct interaction with users is not possible.

In Contribution 8, the authors proposed a facial and emotion detection system using deep-learning techniques and contrast-enhancement image procedures. The authors claim that the proposed system carries 99% accuracy regarding the detection of drivers’ emotions through their real-time facial images. In Contribution 9, the authors proposed a satellite-based object detection framework, which uses SAR images received from a space-based remote sensing installation. This framework operates through solar energy in a lightweight IoT environment with less computational complexity, low power and high precision, so it can help to design large-scale, SAR-image-based, echo-friendly satellite constellations. In Contribution 10, the authors proposed an authentication method using an electrocardiogram (ECG) to protect against forgery attacks. This method uses a binarized ECG signal through the feature-extraction method using artificial neural networks (ANN) to improve the authentication process. The authors report that their proposed system improved the accuracy to 98%, while the error rate is reduced to 2% compared to prior systems.

In Contribution 11, another lightweight cryptographic authentication scheme is proposed for a wireless body area network. The formal proof regarding key management and mutual authentication shows that this scheme has lightweight overheads in terms of energy, computation and communication costs, which is of interest for sustainable systems. In Contributions 12 and 13, the authors proposed other interesting applications of ML methods related to expert opinion and information security. In Contribution 14, the authors proposed a cyber–physical platform for the home appliance recognition and energy management of 5G (and 6G) communication infrastructure. In this article, a juxtaposed analysis is provided of the technologies, standards, and systems that are used. The performance analysis offers a recognition accuracy of 99% for home appliances compared with other related proposals.

In Contribution 15, the authors proposed a simulation model to forecast the spread of the COVID-19 pandemic. The proposed model is derived from the Gaussian curve-fitting method using real data of the pandemic’s evolution in Saudi Arabia. Social distancing measures have changed our lives, especially in densely populated cosmopolitan cities. Therefore, it is of great significance to accurately predict the Coronavirus pandemic using ML- and AI-based methods. In Contribution 16, the authors used ML methods to design intelligent systems for predicting dengue and cancer patients. In Contribution 17, the authors addressed the often-overlooked security and privacy issues associated with the medical IoT. In Contribution 18, the authors proposed a reliable mechanism for remote pain monitoring over wireless body area networks, which uses a trust management system to ensure the authenticity of the parties involved in the communication. The safety rate of the proposed system is reported to be 96%, which helps in the design of sustainable corporate systems. In Contributions 19–21, we found relevant applications of ML methods for the early classification, tracking, and prediction of diseases.

The world must evolve to provide equitable access to unstructured big data from all corporate networks, including energy, transportation, information security, healthcare, and environmental sustainability. This shift requires smart solutions and sustainable networks to improve our living standards. The 21 articles in this Special Issue (listed in the following) demonstrate how modern applications of ML, AI, and IoT can help design useful systems that are increasingly important to our lives, health and well-being.