Differential Assessment of Strategies to Increase Milk Yield in Small-Scale Dairy Farming Systems Using Multi-Agent Modelling and Simulation

Abstract

Multi-agent-based modelling and simulation provides an adequate environment to study the real world. This paper presents the use of a multi-agent research and simulation (MARS) framework and model design based on the overview, design concepts, design (ODD) protocol to model and simulate small-scale management strategies that are important for increased milk yield per cow. In reality, strategies for farm management at a small-scale level are purely based on heuristics that cost farmers and lead to inadequate milk yields. A differential assessment of the farming strategies was conducted to yield a data-driven approach for selection of the best strategies, which in turn will optimize investments and increase milk yield. The agent-based modelling and simulation revealed that, the studied strategies based on income, farm, and farmer-based characteristics influenced an increase of up to 7.72 L of milk above the average (12.7 ± 4.89). Generally, there was an increase in milk yield based on the identified evolvement strategies; from a baseline data average milk yield of 12.7 ± 4.89 to simulated milk yield average of 17.57 ± 0.72. Evaluating the agent-based models in real-world scenarios will strengthen the assurance that the identified strategies can move small-scale dairy farmers from low to higher milk producers.

1. Introduction

Small-scale dairy farming, which is typified by 1–2 milking cows within small herds of 2–5 cattle, provides support for more than 150 million farm households globally [1]. Small-scale farmers provide near to half of the total cattle production in Africa’s livestock farming endeavour [2]. Furthermore, 1.67% of the world’s cattle are found in Tanzania, which has 24 million head of cattle [3]. For many rural and peri-urban people in Sub-Saharan African nations, keeping livestock continues to be essential to their means of subsistence [4,5]. For instance, 50% of people in Tanzania raise livestock to support their livelihoods [6]; and the livestock sector contributes by around one-third of the entire value of agriculture [7]. Despite making a substantial contribution to milk production and meeting market demand, small-scale dairy farmers encounter obstacles that reduce productivity [8]. Small-scale dairy farming initiatives are a vital source of daily sustenance for the majority of farmers who are actively engaged in farming [1].

Small-scale dairy farming is a key agricultural activity in many developing nations due to the valuable food products it generates that enhance household nutrition and food security [9]. Tanzanian farmers have small farms (less than 2 hectares) and cattle herds of one to five cows, and they have low market preference, which means that their products do not command a high price on the market [1]. Due to the increasing demand for milk and dairy products brought on by the world population’s growth to over 7 billion people, the dairy industry employs many people globally and is steadily growing [10]. Nevertheless, despite the anticipated rise in demand for milk and other livestock products in developing nations, the sector continues to confront significant obstacles, such as inadequate infrastructure and the creation and diffusion of new technologies [11].

The increase in the number of people who consume dairy products from 1.8 billion 2009 to 4.9 billion in 2030 predicts a rise in dairy industry but not with the rate of demand [12]. Research initiatives on how to boost milk yield for the setting of small-scale farmers is drawn in by the ever-increasing demand for milk and dairy products [8]. The majority of dairy farming is conducted by crop and animal farmers on small- to medium-sized farms, and it is well known that small-scale producers generate the majority of the world’s dairy [13]. Inadequate dairy production system characterisation and a lack of understanding of the mechanisms influencing their expansion are two challenges [8]. Small-scale dairy producers became more congregated as a result, and there are not as many interventions designed specifically for different farm types [8]. The dairy value chain contributed to this scenario from farm characteristics, service providers, to market systems [14]. According to several experts, one of the main obstacles impeding the sector’s growth is the feeding [15], along with an imbalance in the breed types that are best suited for the various production situations [16]. Improvements in management are required in the areas of feeding, breeding, and disease control if productivity is to grow [17].

This paper presents a multi-agent-based approach for differential assessment of strategies for high milk yield in small-scale dairy farming systems.

The proposed strategy offers a data-driven technique for almost realistic improvement that can be evaluated on the ground and is designed to examine the effects of various tactics for more milk. This is so because the simulation data used is based on actual farming practices and figures for milk yield. Past studies showed how useful multi-agent-based modelling and simulation can be for understanding dynamic, highly interacting systems. Small-scale farming is one such system. According to studies by [8,17,18,19], employing multi-agent-based modelling and simulation to explore farming systems and human behavioral logic is shown to produce positive findings. The studies place a high value on the agents’ improvement-oriented decision making.

On the premise that other parameters affecting farmer and animal behavior and activities are constant during modelling and simulation, the provided data-driven modelling and simulation technique may be constrained. Given the desired role of employing multi-agent-based models to mimic the specifics of a real process, this approach is very delicate. As also employed by [20], face validation approaches are the fastest ways of validating agent-based models. This validity assessment can be complemented by longitudinal and objective approaches [21].

Agent-Based Modelling of Farming Systems

Modelling goals, approach, and data used were the key informants towards understanding how previous research was conducted in the area of livestock farming. A significant body of literature indicates that feeding systems (especially grazing) are widely studied with respect to their effect on livestock productivity and upkeep of vegetation cover. The effects of various policy implementations to individual farmers is equally represented by use of agent-based modelling and simulation.

A study in [22] used agent-based modelling to model and simulate evolvement of a primitive agriculture society based on a single settlement with a heterogeneous landscape that supports agriculture. Agents in this implementation evolved by having production and consumption plans; in which an agent would have inheritors if it produced enough to have surplus after selling and consuming. Agents in the setup needed to learn from their experiences to improve their planning. This knowledge acquisition and transfer (inheritance) presents how the real world evolves. The weakness here is that the evolvement is presented to cover farmers only whilst, in the real world, the given scenario is affected by a number of actors, including humans or systems that support agriculture. However, the authors recommend that inclusion of all entities, as in the real world, it would result into a highly complex model.

In another study, an overview of computational modelling that was meant to address situations in which modelling assumptions are based on fixed neighborhood conditions was presented [23]. Having more realistic and sophisticated models brings in the challenge of solving them analytically. This challenge calls in what the authors called computational economics, in which, complex realistic and sophisticated models need to go through numeric optimisation (rationalisation) and use of simulation methods, in which agents are highly heterogeneous and the whole system is out of equilibrium. Without the use of methods such as agent-based modelling, demand and supply in agriculture-oriented systems may reach an equilibrium if all neighborhood conditions remain fixed. The reality is that, spatial relationships among producers and consumers may influence their demand and supply functions. Therefore, the challenge of model complexity resulting from the heterogeneous nature of the environment can be mitigated by rationalisation, whereby the agents will only adopt association rules that will increase their utility values [23,24].

An agent-based model simulation of long-term climate–livestock and vegetation interactions on communal rangelands conducted by [25] explains the possible outcomes of overgrazing, with climate playing an important role in vegetation growth. The model was developed in order to find critical conditions that can occur in communal rangelands and to suggest other livestock management strategies to farmers if needed. Livestock events in the agent-based model were given as regression equations. To best reflect livestock growth in the real world, age-based categories were implemented for the livestock life cycles.

Similar work in rangeland management is reported by [26], where the model integrated the movement and feeding metabolism of domesticated ruminants. The main goal of the model was to assess the potential of adaptive livestock production in a highly dynamic, heterogeneous, and semi-arid rangeland. Forage selection was highly based on quality and spatial distribution. Water sources were also modelled as individual agents that are determined by climatic conditions. The livestock productivity was modelled in view of forage consumption, conversion of the forage into energy that defined herd fitness.

Productivity of the Ankole–Friesian cattle was modelled in a stochastic simulation model of an Ankole pastoral production system by [27]. A stochastic compartmental model was developed, and the key components were: forage production, herd structure dynamics, and gross margins. Milk production was a sub component in gross margin. Not based on real data, model equations were used to define livestock production.

Understanding livestock farmers’ behavior and adaptation to various strategies was well studied using real data. Authors in [28] described an agent-based model for pastoral farmers’ decision and behavior in response to changes in their operating environment. The main goal was to study production intensity based on the farmers’ choices. On the model, a farmer can learn from fellow farmers and adopt those practices that are delivering better outcomes. At the end, it was identified whether farmers’ networks have a big effect in small-scale or large-scale productivity. Real farm data were used in the model development and simulations. Individual farmer modelling based on real data is also reported by [29], where the response to policy changes was studied with respect to farming intensity.

Authors in [18] conducted a study on farm compliance costs and the nitrogen surplus reduction in mixed dairy farms under grassland-based feeding. Animal and land use activities were simulated under a scenario that all farmers accepted the grassland feeding system. Regression models were then used to predict the reduction in nitrogen emissions for various marginal costs compliance.

Additionally, it was noted that livestock grazing and feeding management is well researched with the use of agent-based modelling. Some agent-based modelling works in farming lacked validation based on real farm data. Although such models might have a wide range of applicability, challenges may arise when considerations are put on the actual situations of the farmers and their dynamics. Farmers from the same place are not necessarily the same or facing similar constraints. Appreciating the distribution and categories of the farmers together with their attributes is important for realistic evolvement models. The progresses reported by [18,28,29] represents how well real farm data can be used. However, farmers’ learning dynamics and adaptation to better practices is presented well in [28]. With the goal of increasing milk yield, this research therefore studied how the farmers can learn from each other and adopt better practices without infringing their social–economic status. In addition, agent-based modelling was applied to study the effect of various farming practices in milk yield.

Unlike other computer modelling approaches that focus on systems behavior, agent-based modelling focuses on individuals’ behavior and their effects on the system being studied [30]. Generalisation cannot be assumed in agent-based modelling since individual agent behavior and attributes will have a great influence on model results. This implies that, in adoption of agent-based modelling for the case of dairy farming, the differences, such as in cattle breeds and production environments, will have a high influence on the model accuracy and applicability. Therefore, a huge limitation towards the adoption of agent-based modelling for dairy farming is the presence of multiple breeds of cattle with varying behaviors, which may hinder a generalised study.

Based on the reviewed studies, the opportunity for the use of real farm data in model development and simulations and application of farm level data in simulation of the real-world scenarios may lower uncertainties that may arise when the models are subjected in real-world evaluation, in addition to reducing heuristics while deciding on optimal farming strategies. With real farm-level data, models can also be developed in a more holistic manner, covering important factors as portrayed in the real world.

2. Materials and Methods

Secondary datasets used in this study were collected through the Program for Emerging Agricultural Research Leaders (PEARL) project, in which small-scale dairy farmers were involved in the baseline survey with an informed consent. Participation in the study was voluntary and every participant signed an informed consent form. The multi-agent research and simulation (MARS) framework [31] was used in the model development and simulations. The framework provides a high scalability of agents in addition to defining agents based on real-world data. The model definition is documented following the ODD protocol [32].

The study adopts a previous assessment of small-scale dairy farming systems in East Africa, where clustering was conducted to establish farm types. Important evolvement strategies for the dairy systems were grouped into Farm, Farmer, Income, and Infrastructure, as detailed in

Table 1. Evolvement strategies for small-scale dairy farm types.

	Farm	Farmer	Income	Infrastructure
Tanzania	Frequency for vaccination	Years of experience	Retail milk sale	Distance to buyers
	Frequency for watering	Years of schooling	Bulk milk sale	Distance to water sources
	Total land	Formal training	Total crop sale	Extension visits per year
	Land for fodder	Membership in groups		Distance to breeding
	Stall feeding in rainy			Distance to market
	Stall feeding in dry
	Grazing in rainy
	Grazing in dry
	Milking cows
	Herd size
	Employed labour
	Months of purchased fodder
	Breeding method

[2,8,11,14].

Table 2. Range of values for numeric features used in the study.

S/No	Feature Name	Type	Range
1	Grazing in dry season	Boolean	0 (no) or 1 (yes)
2	Grazing in rainy season	Boolean	0 (no) or 1 (yes)
3	Stall feeding in dry season	Boolean	0 (no) or 1 (yes)
4	Stall feeding in rainy season	Boolean	0 (no) or 1 (yes)
5	Frequency for watering/day	Discrete	0–4
6	Distance to water source (km)	Continuous	0–15
7	Total land	Continuous	0–100
8	Land for fodder	Continuous	0–80
9	Number of employees	Discrete	1–10
10	Frequency for vaccination/year	Discrete	0–6
11	Membership in groups	Discrete	0–5
12	Years of experience	Discrete	1–50
13	Years of schooling	Discrete	0–21
14	Breeding method	Boolean	0 (bull) or 1 (AI)
15	Distance to breeding (km)	Continuous	0–100
16	Frequency of extension visits/year	Discrete	1–54
17	Herd size	Discrete	1–50
18	Milking cows	Discrete	1–20
19	Peak milk production/day	Continuous	1–40
20	Bulk milk sale (lt)	Continuous	1–100
21	Retail milk sale (lt)	Continuous	1–100
22	Distance to buyers (km)	Continuous	1–37
23	Total crop sale	Continuous	0–950,000 (Tsh)
24	Distance to market (km)	Continuous	0–8
25	Months of purchased fodder	Discrete	1–12

summarizes the limits of numerical features as used in the study.

2.1. Average Milk Yield Model

a. 

Design concepts

• Agent cow: The agent required an amount of energy daily to achieve its potential production. The only source of feeds for the agent was assumed to be grass. Amount of grass eaten by the agent is converted to energy. Daily energy requirements and rates of conversion of grass into energy were obtained from literature [33,34]. The agent’s energy use was allocated for movement, maintenance, resting, milk production, and gain in weight [35]. The model design considered the effects of infectious diseases that can affect milk yield. In order to minimize complexity, a random distribution of clinical mastitis was adopted [36]. By calculating the total probability of catching the infection, the agent cow eating and drinking rates were determined;
• Agent observer: This agent controls changes in the layers, which in turn affects the behaviour of the cow and farmer agents. The observer agent controls grass growth rate and the distribution of mastitis base probability in rainy season and reduces in dry season;
• Agent farmer: This is the farm manager, and is responsible to fill feeds and water troughs for the cows. This agent inputs the dataset that defines how the farm is managed. A change in the input data for the farmer has an impact on the farm productivity, including feeding pattern and annual milk yield from the cow agents. It is the farmer’s set of choices and strategies that is altered to observe changes in the milk yield. The alteration is conducted by adjusting values of the evolvement strategies given in Table 1;
• Adaptation: The cow agent decision room had to decide when to eat and drink based on energy, hunger, and thirsty values. The cow could move to the feeds and water troughs every time either its energy value was below the minimum requirement, hungry or thirsty, observing its food value, current energy level, and hydration in respect to amount of food eaten. Since these values were stored in the cow’s memory, their alteration alerted an action by the cow.
• Objectives: The farmers’ main objective was to increase the amount of milk, which is measured in kilograms per year. The changes in daily milk yield were affected by the choices of management strategies indicated in Table 1; where different values were tested. The cows’ main objective was to increase total energy value measured in MJ to achieve daily requirements and milk yield;
• Interactions: Direct interactions were between individual cow and grass biomass layer, individual cow and feeds and water trough layer, observer and grass biomass layer, farmer and feed and water trough layer, and observer and mastitis layer. Indirect interactions were between individual cows competing for grass and water as they strived to reach their daily objective;
• Stochasticity: Increase in grass biomass, feed and water trough refilling, and distribution of probability for infectious mastitis in the environment;
• Observations: At each simulation time, observations were made on the cow agent (milk level, body weight, and probability of infection), grass layer (increase in biomass), feeds and water trough layer (increase in feeds and water level), and the mastitis layer (random growth and decline of infection probabilities).

b. 

Initialisation

• Cow agent: for each cow agent, the following were the initial values as gathered from existing research [33,34]:

  ○

  Location: randomly distributed in 45 × 45 grid cells;

  ○

  Instances: 5;

  ○

  Energy (MJ) = 0;

  ○

  Milk level (kg) = 0.5;

  ○

  Probability of infection = ((0.16 × Mhistory)/5) + ((0.16 × parity)/5) + (0.16 × (1/breed));

  ○

  Body weight (kg) = initial value given for each breed in an input data file;

  ○

  Breed = fixed breed type given for each agent in an input data file;

  ○

  Mastitis history = initial value given for each agent in an input data file;

  ○

  Parity = initial value given for each agent in an input data file;

  ○

  Farmer agent: the farmer agent was randomly distributed in a 45 × 45 grid cells, and a set of values for each evolvement determinant (as indicated in Table 1) was given as an input file;

  ○

  Observer agent: for the observer agent, initial values were a random distribution in a 45 × 45 grid cells, and one instance.

c. 

Model layers

The implementation of layers was used to have highly flexible model interactions and improve focus on the individual agent [31]. Figure 1 shows the organisation of three independent layers, in which the agents interact with each other.

• Grass layer: Each cell in a 100 × 100 grid contained 4 kg of grass at initialisation and increases/reduces randomly during simulation;
• Feeds and water trough: 4 × 9 cells for feeds and 9 × 9 cells for water out of 100 × 100 grid. The farmer agent randomly increased values of the cells from 0 (at initialisation) depending on management practices given in the farmer input file;
• Mastitis base probability: Each cell in a 100 × 100 grid had 0 base probability at initialisation and increased or reduced randomly during simulation.

d. 

Input data

Input data were generated for the model layers, cow, and farmer agents. The raster files for the layers were in .asc format and each cell carried a value of 4 for the grass raster, and 0 for the feeds and water trough, and the mastitis raster files.

• Grass raster file (100 × 100);
• Feeds and water trough raster file (100 × 100);
• Mastitis probability raster file (100 × 100);
• Cow agent input csv file;
• Farmer agent input csv file.

Figure 1. Average milk yield model layers.

Figure 1. Average milk yield model layers.

2.2. Milk Yield

Unlike in the farmer networks modelling, where average milk yield estimation model was derived by a regression model, in this milk yield model, the cows are left to produce milk as an effect of their eating patterns, energy requirements, and as affected by the farmer’s decisions. The following mathematical expressions (Equations (1)–(6)) were used to calculate the daily energy requirements to milk yield and increase in body weight (for individual cow agents).

• Energy requirements (equation adapted from [37])

$$EM=6.9\left(42.4B_w^{0.75}+442M_y\right)\left(1+M_y-15\right)0.00165$$

(1)

where EM is maximum daily energy requirement, B_w is body weight, and M_y is maximum milk yield.

• Derived probability of infection

$$P\left(I\right)=P_e+0.0032a+0.0032b+\frac{0.016}{c}$$

(2)

where P(I) is total probability of infection, P_e is probability of infection from the external environment, a is mastitis history, b is parity, and c is breed type.

• Derived conversion of grass matter and water to milk yield in accordance to [38,39].

$$y=0.2f+0.0625w$$

(3)

where y is milk yield, f is food value, and w is water intake.

• Derived energy use

$$E_u=\left(10+0.1B_w\right)+7.1y+20$$

(4)

where E_u is daily energy use, B_w is body weight, and y is milk yield. The constants 10 and 20 were used for the assumptions that 10 MJ is a standard amount of energy required for body maintenance for the cows (the total amount will change depending on the cow’s body weight), and 20 MJ is a standard amount of energy required for movements for the cows.

• Derived excess energy

$$E_e=EM-E_u.$$

(5)

• Derived weight gain

$$B_w=\frac{E_e}{44}.$$

(6)

2.3. Simulation Scenarios

• To understand the effect of the evolvement strategies (Table 1) on average annual milk yield;
• To identify among the categories of evolvement strategies (farm, farmer, income, and infrastructure characteristics), which category has a higher likelihood of improving milk yield.

3. Results

3.1. Assessment of Farm Characteristics

Simulation results indicate that farmers use purchased fodder get an increase in their average milk yield. Therefore, lack of purchased fodder led to 28% drop in milk yield from 22 L to 16 L. An increase in the level of employed labours and watering also resulted into an increase in milk yield, while an increase in herd size, number of milking cows, use of grazing, and total land caused up to a 11.77% decrease in milk yield (Figure 2). Other farm characteristics, such as vaccination and choice between stall feeding and mixed did not make a significant change in milk yield. The farm characteristics significantly explained the variances in milk yield, with an r² of 0.96 when combined.

3.2. Assessment of Farmer Characteristics

Experience and formal training on dairy care were shown to have a positive impact on milk yield, yielding up to 18 L of milk, equivalent to an increase of 4.3 L above the average yield in actual values. There was a slight difference in milk yield based on years of formal schooling and membership in farmer groups (Figure 3). The farmer characteristics alone could explain a significant level of variance in the milk yield (r² = 0.94).

3.3. Assessment of Income Characteristics

Regular sales of milk were associated with an increase of 4.8 L of milk. Sale of bulk milk indicated a drop of 4.5% from farmers who had less sales of bulk milk. Figure 4 indicates a drop of milk yield from 18 L to approximately 17.3 L for farmers who sell milk in bulk. When the three characteristics were fitted, a r² of 0.93 was obtained as a proportion of the variance explained by the income characteristics.

3.4. Assessment of Infrastructure Characteristics

Farmers located far from improved breeding services demonstrated an increase of 5.8 L above the average for the actual data, giving a milk yield of 18.5 L more than those living near service providers (16.5 L). On the other hand, short distances to milk buyers, water sources, and at least nine extension visits per year resulted to an increase of at least 4.3 L of milk above the average for actual yield (Figure 5). Moreover, farmers living near formal markets had lower milk yield than their peers by at least 0.2 L. Considering the effect of sales given in Figure 4, more milk sales implied more milk yield; it is becoming evident that farmers do not sell their milk in formal markets. When all the infrastructure characteristics were fitted, an r² of 0.93 was obtained.

The simulation results for the milk yield model indicate the differential influence of each evolvement determinant towards increase in milk yield for Tanzania case studies. The differences in milk yield that is L of milk above or below the average for actual yield was considered in ordering the evolvement strategies in addition to the r² obtained in each model run. Consequently, for Tanzania; farm, farmer, income, then infrastructure characteristics was the order of importance. Table 1 lists the evolvement strategies.

4. Discussion

The simulation results conclude that, in order of priority, if farmers from Tanzania’s small-scale dairy system have to increase their milk yield, then; farm, farmer, income, and infrastructure characteristics have to be considered as given in Figure 2, Figure 3, Figure 4 and Figure 5. The agent-based modelling and simulation revealed the studied strategies based on income, farm, and farmer-based characteristics influenced an increase of up to 7.72 L of milk above the average (12.7 ± 4.89). The identified strategies could predict up to 96% of the variances in milk yield for Tanzania.

The presented findings fall in agreement with other research on the role of income generation, strategies for farm management, and farmer’s education and experience in sustainable dairy production. In view of [40], income and social economic factors demonstrated significant influence on dairy production, which can imply that farmers are more capable to invest into sustainable dairy production. The main sources of income for dairy producers are milk and crop sales (for mixed production systems). In the study of [41], income and formal credits have positive influence on dairy production. These studies further clarify the need for financial security and a stable socioeconomic status for the farmers to be able to afford various high-yielding production strategies.

Studies indicate the importance of stable feeds production in addition to supplement feeding for dairy cow nutrient requirements [42]. Furthermore, Refs. [42,43] affirm the importance of high-yielding farm management strategies, including supplement feeding for additional nutrients, balance of feeds, and adequate watering frequency.

Among the observed farm characteristics that can directly be related to income are the ability of farmers to purchase fodder, available land for fodder production, manageable herds, mixed feeding approaches during wet and dry seasons, water sources, and animal vaccination. Therefore, income sources can largely influence the success of such practices, as they could require access to formal credits that can support production [41].

Furthermore, presented findings denote a slight increase in average milk yield for farmers with at least secondary school education. In other studies, education and experience were also linked to better dairy performance by small-scale farmers and their influence on choosing best farm management practices. As shown by the example in [44], experienced and educated farmers could better support vaccination programs.

Generally, there was an increase in milk yield based on the identified evolvement strategies; from the baseline data average milk yield of 12.7 ± 4.89 to simulated milk yield average of 17.57 ± 0.72. Evaluating the agent-based models in real-world scenarios will strengthen the assurance that the identified strategies can move small-scale dairy farmers from low to higher milk yield. This can be achieved by deploying a self-assessment model based on mobile applications that farmers can interact with on a daily basis as suggested in [2]. As such, in cases of limited resources, farmers should not go through a trial process to identify a working strategy. To our knowledge, this is the first study that explored evolvement strategies for farmers in Tanzania by using unsupervised learning approaches, agent-based modelling, and simulation. The reported experiments, which are based on artificial intelligence, complement the projections reported for the development of the dairy sector by 2067 as studied in previous research [45].

The presented approach to study the effects of various strategies for more milk provides a data driven technique for near to realistic improvement that can be studied on the ground, this is because the used simulation data are based on actual strategies and milk yield values as collected from farmers. Previous research demonstrated the potentials of agent-based modelling and simulation in studying dynamic and highly interactive systems, of which small-scale farming is an example. The studies of [8,18,19,29,31] outline promising results for using agent-based modelling and simulation in studying farming systems and human behavioral logic. The studies highly attribute to the agents’ decision making towards improvement paradigms of the farming systems. However, the presented data-driven modelling and simulation technique can be limited with an assumption that, during modelling and simulation, other factors around farmer, animal behavior, and activities are constant. This is particularly a sensitive notion given the ideal role of using agent-based models to imitate the details of a mechanism or real process. Face validation techniques are the quickest methods of validation, as also used in [20]. For an objective validation of the presented study findings, a longitudinal study could complement the validity assessment as exemplified in [21].

5. Conclusions

In order to enhance on-farm decision making for small-scale dairy farmers, an assessment of important strategies for an increase in productivity is required. The presented study provides a differential assessment of various farm strategies and their likelihood of yielding more milk per cow. Preliminary results indicate significant increase in milk yield. A thorough validation of these strategies is proposed to affirm the results. Future research is proposed in yielding this assessment on a farmer-centric mobile application to enable on-farm assessment.