Weather variable selection for whitefly population prediction modeling by using backward elimination regression

The present investigation discusses the selection process of the most influencing weather variables for developing a prediction model for whitefly, Bemisia tabaci (Gennadius), based on the backward elimination method. This method aids in the selection of a model with fewer variables by eliminating those that are less pertinent, thereby enhancing precision and mitigating model complexity. In the pursuit of achieving a balance between simplicity and model fit, the conventional 5% level of significance (p-value ≤ 0.05) was utilized along with six weather variables viz., maximum temperature, minimum temperature, evaporation rate, sunshine hours, rainfall, and evening relative humidity. Through an iterative elimination process, it was determined that only three variables-minimum temperature, sunshine hours, and evening relative humidity-significantly contributed to the prediction model. Subsequently, these three variables were retained for predicting whitefly population counts, while the remaining less relevant variables were discarded. The model was found to be around 74 percent accurate in predicting the dynamics of whitefly.