Machine learning in environmental sustainability factor analysis in the agricultural sector

¹ Moscow Timiryazev Agricultural Academy, Russian State Agrarian University, 127550 Moscow, Russia
² Reshetnev Siberian State University of Science and Technology, 660037 Krasnoyarsk, Russia
³ Bauman Moscow State Technical University, 105005 Moscow, Russia

^* Corresponding author: ankoz9@yandex.ru

Abstract

The study employed several key data analysis methods aimed at enhancing the understanding of relationships between variables and improving prediction accuracy. The primary tool used was correlation analysis, which allowed for the identification of the degree of association between two variables by determining how changes in one variable relate to changes in another. This established a foundation for further in-depth data analysis. For a deeper understanding and simplified interpretation of the data, factor analysis was utilized. This method helped to identify latent factors that explain the relationships between observed variables and to reduce the number of variables by grouping them. This made the analysis easier and facilitated the identification of key components affecting the data. Logistic regression was applied to build data models. This method is used to model the probability of a specific event occurring based on independent variables, allowing for the classification and prediction of categorical outcomes. The logistic function was used to estimate probabilities and the relationship between the dependent variable and predictors. To enhance the performance of the logistic regression model, a Weight of Evidence (WoE) analysis was conducted. This method converts categorical and continuous variables into numerical formats, simplifying data interpretation and improving the model’s predictive capabilities. WoE analysis helps to identify significant factors, improve the linear relationship between predictors and the dependent variable, and reduce the impact of outliers, which is particularly important in areas such as credit scoring. The results of applying these methods showed that the model based on correlation and factor analysis explained 27.51% of the information on the training set and 76.04% on the test set.