Rainfall Prediction Using Data Mining Techniques: A Performance Analysis

Abstract

Rainfall prediction is a crucial aspect of weather forecasting, as accurate and timely predictions enable the implementation of effective precautionary measures across various sectors, including transportation, agriculture, construction, flight operations, and flood management. By leveraging historical data, data mining techniques offer a promising approach to predicting rainfall based on key features. This research presents a critical analysis of data mining techniques employed for rainfall prediction, highlighting their strengths and limitations. The analysis reveals that while these techniques show promising results in classifying non-rain conditions, they perform poorly in accurately predicting rainfall events. The Support Vector Machine algorithm, for instance, achieved an F-measure of 0.958 for non-rain classification, but failed to predict rainfall with an F-measure of 0. Similarly, the Random Forest Method exhibited strong performance in classifying non-rain events, with an F-measure of 0.946, but only managed an F-measure of 0.357 for rainfall prediction. These results suggest that the data mining models struggled with rainfall classification due to various factors, including missing values, the absence of critical climate attributes, and the dataset’s lack of quantitative rainfall measurements. Moreover, the dataset's focus on categorical weather conditions, rather than specific rainfall amounts, further limited the patterns available for classification. This work provides valuable insights into the limitations of current data mining approaches and establishes a foundation for future research aimed at improving rainfall prediction accuracy.