Logistic Regression Based Next-Day Rain Prediction Model

Abstract

Rain prediction is challenging due to the complex combination of atmospheric factors. This paper presents the application of logistic regression modelling to predict rainfall the next day, using weather parameters from previous days. One year of weather data (temperature, pressure, humidity, sunshine, evaporation, cloud cover, wind direction, and wind speed) from Canberra, Australia, has been used to develop the logistic regression model. The best fit logistic regression models are selected using Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and Least Absolute Shrinkage and Selection Operator (LASSO) based on feature selection for predictive modelling. These models are evaluated using Area Under the receiver operating characteristics Curve (AUC) and Hosmer-Lemeshow test to determine each model's accuracy and goodness of fit to predict rainfall occurrence the next day. The likelihood of rainfall the next day has been interpreted based on the calculated odds of the selected independent weather parameters. The result shows that using logistic regression (AIC Backward), rainfall the next day can be predicted with 87% accuracy, provided that the appropriate weather parameters are chosen.