1. Introduction
A crucial part of human activity throughout history has been related to providing the right amount of food. Although humanity has made significant progress in this regard in contemporary times, the concept of food security identified at the World Food Conference in Rome in 1974 and developed at World Food Summits in 1996 and 2009 still receives considerable attention from policymakers at the global and national levels, as hunger and malnutrition have been not eliminated. The fundamental idea behind food security is that all individuals have economic, social, and physical access to an adequate and nutritious food supply to meet their dietary needs for a stable and active life (
FAO 2009). Such an approach is demanding and ambitious and, in its current formulation, can be considered as a long-term objective for developing and developed countries alike. In the contemporary context, the focus is on the nutritional dimension of food security. In 2015, United Nations member states set up the 2030 Agenda for Sustainable Development, with 17 Sustainable Development Goals (
United Nations 2015). Goal 2 is about creating a world free of hunger by 2030. Achieving this goal appears challenging in light of the FAO’s indicator reflecting the overall level of food security: the prevalence of undernourishment. Globally, this indicator increased from 7.5% in 2017 (the lowest level in the history of the assessment) to 7.9% in 2019 (pre-pandemic) and 8.9% in 2020. The upward trend persisted in 2021, reaching 9.3%, and in 2022 it remained relatively unchanged (
FAO 2023b).
However, this issue varies spatially. In highly developed countries, food insecurity occurs, but the problem is rather distributive; however, it is very harmful to the affected groups. For example, in the United States in 2021, 3.8% (5.1 million) of U.S. households had deficient food security (
Coleman-Jensen et al. 2022). Developing countries suffer from insufficient amounts of food. Hunger and undernourishment are serious problems, particularly in numerous Asian and African countries. According to the
FAO (
2023b), hunger affected 19.7% of the African population and 8.5% of the Asian population in 2021.
Zereyesus et al. (
2022) noted with concern that the number of food-insecure people in 2022 in the 77 low- and middle-income countries covered by their investigation had increased by 9.8% (118.7 million people) from the 2021 estimate. However, this indicator remained relatively low (at 2.5%) in Europe and Central Asia over the past few years. Despite this, 116.3 million people experienced moderate or severe food insecurity in the region, and over the past two years this number increased significantly by 25.5 million people. In Central Asia, 20.2% of the population faced moderate or severe food insecurity (
FAO 2023a).
There has been substantial existing research on food security, both globally (
Godfray et al. 2010;
Stephens et al. 2018;
Lal 2016;
Saboori et al. 2022) and within specific countries (
Lv et al. 2022;
Islam 2014;
Loginov 2020;
Wineman 2016). However, there are notable gaps in our understanding of the specific dynamics of food consumption patterns, particularly within the framework of developing economies such as Kazakhstan. Studies on food consumption in Kazakhstan have predominantly concentrated on specific categories like livestock (
Liang et al. 2020;
Akhmetova et al. 2022) or fruits and vegetables (
Abe et al. 2013), with limited attention to overall food consumption. However, there is a lack of recent research, often missing detailed examinations of socioeconomic variables affecting food consumption, especially in the aftermath of geopolitical conflicts and economic shocks. For instance, previous research has primarily focused on macro-level factors such as agricultural production (
Gizzatova et al. 2014) and imports (
Denissova and Rakhimberdinova 2021), often neglecting consumer-related factors. The recent global events have underscored the necessity for targeted research on the relationships between socioeconomic factors and food security in the region.
Food security is especially vulnerable to risks stemming from shocks in the economy or its environment. Over the period 2000–2022, three major shocks occurred: the global financial crisis of 2008–2009 (resulting in the most significant and sharpest decline in global economic activity of the modern era), the global pandemic of COVID-19, and the Ukrainian–Russian war. The modes and strengths of the influence of each shock differ. According to
Travasso et al. (
2023), all over the world, the 2008 crisis drove 97 million more people compared to the pre-crisis year 2007 into hunger. COVID-19 was more severe, driving 112.6 million more into hunger. Due to the conflict in Ukraine, the number of people who are food-insecure or at high risk has soared to reach a record high of 345 million. Countries that are highly dependent on food imports are particularly susceptible to food insecurity because of this conflict (
Abay et al. 2023). Approximately 50 million people across 45 countries are at risk of falling into famine or famine-like conditions (
WFP 2023a).
Jarrell et al. (
2023) noted that despite international efforts successfully slowing down the hunger crisis in 2008–2009, they did not establish a support system for food security that would prevent the magnitude of today’s (2023 perspective) crisis. This means that neither international organizations nor particular countries have implemented appropriate measures to mitigate the risk of food insecurity stemming from different shocks and create secure, sustainable, and resilient food systems.
The volume and structure of food consumption per capita are the primary indicators of meeting the nutritional needs of individuals. Ensuring food security is associated with two key indicators: the production and consumption of food products (
Rabbi et al. 2021). These are determined by complex factors: economic, political, demographic, and social. The shocks mentioned above can drastically change some factors while leaving others unaffected. At the macro level, regulatory bodies such as governments and state agencies seek to understand their power and direction of impact to encourage structural changes in the food and agriculture industries or respond to challenges (
USDA ERS 2023). The European Food and Safety Authority emphasizes that consumption data are essential for assessing how people are exposed to potential risks in the food chain (
EFSA 2023).
The problem of undernourishment in Kazakhstan was significant in the early 2000s. The undernourishment rate was 6.5%, and the trend was upward. Since 2009, the situation has gradually improved, but the problem still requires a satisfactory resolution. The average rate of moderate or severe food insecurity was 2.1% for 2017–2019, but it increased to 2.7% for 2019–2021, reflecting the impact of the COVID-19 pandemic shock (
FAO 2023a). This is why food security is a paramount goal in agricultural and economic policies in Kazakhstan (
Aimurzina et al. 2018), becoming one of the principal conditions for ensuring national security.
Despite the absence of specific legislation directly addressing food security (
Aigarinova et al. 2014), the Act “About National Security of the Republic of Kazakhstan” recognizes its importance by defining food security as a state of the economy, including the agro-production sector, wherein the government can ensure the physical and economic access of the nation to quality and secure food products, sufficient to meet the physiological consumption rates and accommodate demographic growth (
Parliament of the Republic of Kazakhstan 2023). “The National Plan for Ensuring Food Security of the Republic of Kazakhstan for the Years 2022–2024” (
Republic of Kazakhstan 2022) outlines a comprehensive action strategy encompassing the development of agricultural production, advancement of agrarian science, education, and knowledge dissemination within the agricultural sector, ensuring economic access to food products, and enhancing the quality and safety of food products. Projected outcomes of these steps include the ability to forecast prices for the most notable food products for 1.5–2 months, proposing targeted support for socially vulnerable segments of the population, and assessing the feasibility of establishing stabilization funds. Furthermore, the uneven distribution of the risk of food insecurity among social groups exacerbates the challenge of adequately directed support, as indicated by the FAO, WFP, and other sources (
Rabbitt et al. 2017;
Pool and Dooris 2022;
Grimaccia and Naccarato 2022;
WFP 2023b;
FAO 2023c).
Following the unexpected Russian invasion of Ukraine, which has noticeably impacted Kazakhstan’s economy (
Agaidarov et al. 2023), achieving the objectives outlined in the National Plan has been made challenging. As Kazakhstan is a close economic partner of Russia, internal and external shocks related to Russia, especially sanction measures, have a negative economic effect on Kazakhstan. For instance, rapidly rising inflation poses a serious threat to Kazakhstan as an import-dependent country, resulting in declining living standards, inflationary pressures, and reduced purchasing power due to anti-Russian sanctions and retaliatory measures by the Russian government. Central Asian countries, including Kazakhstan, maintain long-term trade ties with Ukraine, exposing them to secondary impacts such as rising food prices, which generally have a negative impact on the countries’ food affordability and food security. These issues are especially aggravated due to the instability of the national currency.
Existing studies in Kazakhstan, in terms of theoretical methodology, include only the emergy method (
Jia and Zhen 2021), the scenario approach (
Khishauyeva et al. 2017), and the method of integrated indicators (
Stukach et al. 2022). The new approach proposed in this article relies on neural network modelling, specifically tailored to the unique context of Kazakhstan’s food consumption patterns, thereby enhancing our theoretical understanding of how socioeconomic factors intersect with food security issues. Other researchers have explored various combinations of modelling and architecture models to address similar challenges. A neural network is a powerful tool for identifying complex trends within datasets.
Abdella et al. (
2020) leveraged the impact of environmental, economic, and social indicators on the 29 food consumption categories in the USA using k-means clustering and logistic regression, which are machine learning methods. The model’s results showed that the overall accuracy was 91.67%, indicating considerable efficiency. In a different work,
Wang et al. (
2010) used GM (1, 1) grey modelling, BP neural network modelling, and a composite of grey–neural network modelling to predict global food consumption. The best result was demonstrated by the combination model, considering efficiency based on M (mean absolute error), MP (average relative error), and T (Theil inequality coefficient).
Gerber Machado et al. (
2020) employed a neural network based on four independent variables—sex, ethnicity, level of education, and income group (input layer)—with four neurons in the hidden layer, predicting the dependent variable of food consumption per household per capita per day in Brazil (output layer). The value of the regression coefficient R at the level of 0.90958 indicates significant network efficiency.
Alfred et al. (
2022) applied machine learning methods to model the consumption per capita of 33 fresh agro-food items in Malaysia based on the total GDP per capita. A resilient backpropagation neural network was applied in their research. The neural network included three types of models, where the first model, with one input, one hidden layer (10 neurons), and one output, demonstrated the lowest total MSE (mean square error) of 17.95 compared to the other types of models. Considering the consistent evidence from various studies, it is clear that neural networks exhibit high-quality forecasting capabilities. Hence, we utilized them in our research endeavors. Moreover, it is essential to note that these neural networks differ in structure, parameters, and methodologies. Each research problem has its unique characteristics and intricacies. Therefore, it is imperative to treat each research challenge individually, customizing the neural network’s architecture and parameters to best suit the specific dataset and objectives at hand.
Understanding the relationship between factors and food consumption is crucial for formulating effective strategies to promote food security. The recognition of the influence characteristics is a precondition to identifying the risks that they pose and implementing countermeasures. The increase in the population can intensify the pressure on food consumption (
Jia et al. 2023). Household sizes, due to household economies of scale, decrease the food consumption per capita (
Nelson 1988), while household income influences consumption expenditure patterns (
Barigozzi et al. 2012;
Tajaddini and Gholipour 2018). The unemployment rate affects household budgets and, as a result, changes the dietary intake (
Antelo et al. 2017;
Smed et al. 2018) and influences dietary composition and food consumption patterns. Economic factors such as gross domestic product (GDP) (indicating the economic prosperity of a certain country) (
Jia et al. 2023;
Martini et al. 2022) determine the income level and its increase, especially in less developed countries, which can result in increased food demand and changes in its structure (lower income influences the food choice) (
Erokhin et al. 2021). As mainly low-income social groups are susceptible to the risk of food insecurity, the income distribution matters (
Rosen and Shapouri 2001;
Rashidi Chegini et al. 2021). Reducing income inequalities can be expected to contribute to a decline in the risk of food insecurity. The rise in food price inflation negatively affects the food consumption models (
Bozsik et al. 2022;
Green et al. 2013;
Martini et al. 2022). An increase in the poverty rate means a drop in food purchasing power, intensifies threats of undernutrition (thereby increasing the risk of food insecurity) (
Hjelm et al. 2016;
Siddiqui et al. 2020), and reduces people’s food options, impacting the adjustment in the food consumption structure (
Chang et al. 2009). The average subsistence level per capita to some extent corresponds to the poverty rate. Its increase means that consumers have to change their consumption structure. According to
Jensen and Miller (
2008), the elasticity of demand (food) is substantially dependent on the severity of poverty.
Predicting and mitigating the risk of food insecurity is a vital issue amid geopolitical conflicts, pandemics, and economic crises. Accurate predictions of food consumption trends allow governments to allocate resources to the risk-prone areas where shocks affect vulnerable populations. Moreover, government organizations can leverage the findings to formulate targeted interventions and develop strategies that mitigate the impact of shocks on food security. This study aligns with the objectives outlined in the “The National Plan for Ensuring Food Security of the Republic of Kazakhstan for the Years 2022–2024”, providing valuable insights to the government for addressing current and emerging food security challenges.
The main goal of this paper is to develop a neural network predictive model for forecasting food consumption patterns in Kazakhstan, so as to mitigate the risk of food insecurity. The first step of this research was the identification of socioeconomic variables that have a significant impact on food consumption patterns in Kazakhstan. For this purpose, principal component analysis (PCA) formed the basis for the subsequent analysis and made it possible to determine the most critical variables.
Taking into account the analysis of the literature on socioeconomic factors affecting food consumption as input data for the PCA analysis, the following variables were used: population growth rate (%), GDP per capita (KZT), food price and tariff index (previous year = 100), poverty rate (%), income concentration ratio (Gini index), average household size (people), average subsistence level per capita (KZT), and unemployment rate (%).
The main research contributions of this paper are as follows:
Analyzing historical data related to food consumption and economic factors, and identifying dynamic trends and patterns.
Using PCA analysis to indicate the key variables/factors for food consumption and their values for modelling a neural network.
Demonstrating the computational power of an artificial neural network (ANN) to create models capable of successfully predicting total food consumption and the percentage distribution of different food consumption categories.
Establishing the foundational architecture and parameters of the ANN models.
Creating a neural network model that can be used to directly predict total food consumption and the percentage distribution of different food consumption categories, thus presenting an alternative to the existing calculation methods. Thus, the government has another useful method to develop economic policies tailored to current and short-term needs.
4. Discussion
The scientific investigations outlined in this research provide a comprehensive understanding of the use of different machine learning approaches to predict food consumption patterns. These investigations highlight the potential of these methods for accurate forecasts, which can be valuable for shaping policies and strategies for food security.
This study is notable for employing neural networks to predict how various food consumption categories are distributed in Kazakhstan. Moreover, the investigation revealed that the best setup for distributing the various food consumption categories and the overall food consumption involved using a network with eight neurons and nine neurons, respectively. The minimal prediction error and high regression value both for the model for the proportional distribution of various food consumption categories (R = 0.99973) and for the total food consumption model (R = 0.99488) demonstrate the effectiveness of the neural network model in forecasting food consumption trends. High R values approaching 1 indicate a strong correlation between the predicted values and actual observations, emphasizing the robust performance of the model in capturing and predicting complex food consumption patterns.
These findings suggest that neural network modelling could be a valuable tool for predicting food consumption trends, assisting in planning and policymaking for food security.
Examining global research, the investigation into the per capita consumption of several fresh agro-food commodities in Malaysia highlights the effectiveness of neural networks. The study showed that the neural network (NN) model outperformed the ordinary least squares (OLS) model, resulting in a lower aggregate mean squared error (MSE). The NN model demonstrated the lowest total MSE of 17.95 for all 33 fresh agro-foods investigated in this study (
Alfred et al. 2022). This finding further strengthens the argument for using neural networks in forecasting food consumption trends.
Studies by Giulia Martini and associates, Pietro Foini et al., and Deléglise et al. highlight the application of machine learning in predicting food security situations.
Martini et al. (
2022) employed XGBoost, explaining up to 81% of the variation in insufficient food consumption and up to 73% of the variation in crisis or above food-based coping levels. The research by
Foini et al. (
2023) demonstrated that precise forecasts of insufficient food consumption levels could be made up to 30 days into the future, thereby informing decisions regarding the allocation of need-based humanitarian assistance. However,
Deléglise et al. (
2020) discovered that predicting food security indices is a challenging issue; their models did not exceed R
2 = 0.38 for the Household Dietary Diversity Score (HDDS) and R
2 = 0.35 for the Food Consumption Score (FCS).
These findings collectively show the potential of machine learning paradigms, especially neural networks, for predicting food consumption trends. While the successful application of neural networks in forecasting consumption patterns is evident, this research lays a strong foundation for future explorations in this field. The ability of machine learning models, such as neural networks, to clarify differences in food consumption and coping levels opens promising opportunities for further research and progress. These insights contribute to the increasing knowledge about using advanced computational methods to better understand the dynamics of food security.
5. Conclusions
This paper presents a comprehensive study focused on building a neural network model to forecast food consumption trends in Kazakhstan, with the goal of reducing the potential risk of food insecurity. The initial phase of the research concentrated on identifying socioeconomic factors that significantly influence how people consume food in Kazakhstan. To achieve this, we used principal component analysis (PCA) as the main method for further evaluations. PCA facilitated the identification of important factors, such as natural population growth (%), GDP per capita (KZT), food price and tariff index (previous year = 100), poverty rate (%), average household size (people), and average subsistence level per capita (KZT). In addition, considering the PCA results, especially the values of principal components #1 and #2, revealed that the most crucial factors affecting food consumption in Kazakhstan are the poverty rate, GDP per capita, and food price index.
The results of the PCA analysis were the input variables used to build the ANN models. Two models were built: to predict food consumption on a national scale per capita per month, and to model the percentage distribution of various food consumption categories. The general regression for the first network was 0.99488, while that for the second was 0.9973.
The forecast results show a low prediction error of less than 10%, signifying a high level of accuracy in the model’s predictions. A low prediction error is crucial in forecasting, as it indicates that the model’s estimates closely align with the actual values. Overall, the positive indicators in modelling quality support the idea that network modelling can predict total food consumption (per capita per month) in Kazakhstan and the percentage distribution of different food consumption categories in Kazakhstan. Examining the forecasted total food consumption for the next three years suggests a decrease, influenced by rapid population growth due to Ukrainian–Russian war migration and high inflation. Considering the projected trends in the food consumption structure for the next two years, we anticipate the following: firstly, the proportion of bread consumption is expected to stay relatively stable; secondly, there will be an increase in the consumption of meat, potatoes, fruits, and vegetables, along with a decrease in the consumption of dairy products and oils. Notably, significant deviations from these patterns are predicted to arise in 2025, marked by a decline in potato and meat consumption and an increase in the consumption of bread and dairy products.