Modeling Climate Impact on Agricultural Systems Using Machine Learning Methods

In this research, various models based on machine learning algorithms and neural networks were created to predict air quality. Various regression models were constructed to predict the air quality index (AQI). For the regression task, a comparative analysis of modern gradient boosting algorithms for AQI forecasting, including XGBoost, LightGBM, and CatBoost, was conducted. A comparative analysis of the baseline and optimized models was conducted. Regression curves were plotted, and their performance was analyzed. A forecasting comparison of the performance of neural networks with the Adam and RMSProp optimizers for predicting air quality was conducted. A multivariate regression model was constructed and optimized. To improve model quality, neural network hyperparameters were optimized. A neural network was constructed, and the results were compared with the Adam and RMSProp optimizers. Their accuracy and model error were assessed, and error matrices and ROC curves were constructed, showing air quality. Using the Random Forest model, we predicted the concentration of AQI (or other particulate matter) based on data on PM2, 5, PM10, CH4, CO2, NO2, SO2, pollutants, meteorological parameters (temperature, humidity, wind speed, etc.), and temporal variables. Nonlinear and hidden relationships between the variables and AQI were identified using the traditional Pearson correlation matrix when the data are normally distributed, and Spearman and Kendall rank correlation matrices when the data are not normally distributed. Temporal patterns of AQI autocorrelation and cross-correlation with meteorological parameters were determined.