Application of machine learning models for predicting whitefly population on tomato under greenhouse conditions

Climatic conditions such as temperature, relative humidity, and radiation are important factors regulating insect infestation under natural and greenhouse conditions. The whitefly is one of the most important insect pests affecting the plant yields of major crops. Whitefly infestation of tomato plants is a common occurrence under greenhouse conditions. In this study, climatic conditions such as outdoor temperature (A), greenhouse internal temperature (B), relative humidity (%RH-C), HD (D), and radiation (E) were recorded weekly for 1 year (52 weeks) using water sensors. Greenhouse-mounted traps were used for whitefly counts every week. The results obtained were subjected to factorial regression analysis using the Minitab statistical program. The results were also subjected to different statistical analyses such as Pareto charts, normal graph analysis, contour plots, and surface plots. When the Pareto chart results were analyzed, it was found that all factors (ABCDE) were effective at the top. Normal plot analysis showed that all input factors are distributed close to the fitted line on both left and right sides. The results of contour plots and surface plots showed the data to be distributed between -5000 and 5000 whiteflies per week. Optimization tools calculated that there would be 500 whiteflies at low radiation if the whitefly infestation conditions were optimized. The data was also validated and predicted using three different machine learning models with the Quasi-Poisson model. The model's performance is six different. The R2 results show that XGBoost (0.739) has the best performance, followed closely by RF (0.737) and MLP (0.677) models, while RMSE shows a similar performance. The order for MAE, MAPE, and MSLE metrics was RF-XGboost-MLP. In general, both RF and XGBoost models outperformed the MLP model for both actual and predicted whitefly numbers. Heat map analysis showed a positive but variable correlation between climatic conditions and whitefly infestation. It is concluded that statistical tools can be successfully used to optimize any numerical data related to insect infestation.