Designing a rule-based hourly rainfall prediction model

Abstract

Rainfall prediction is important in many aspects of our economy and general livelihood by preventing any serious natural disasters. Although numerous methodologies have been introduced to predict hourly rainfall, most of them cannot provide transparency of predicted outcomes. To overcome this limitation, we propose a computer-aided rule-based rainfall prediction model using CART and C4.5. To correctly perform rainfall prediction, the chance of rain is first determined. Then, hourly rainfall prediction is performed only if there is any chance of rain. As outcomes, rules for rainfall prediction are provided, which may provide hidden, but important patterns with transparent reasons. To validate the rules, ten-fold cross validation method is performed and its average accuracy with precision and recall is reported. As results, the average accuracies of classifying and predicting raining conditions with CART and C4.5 are 99.2% and 99.3%, respectively. And the average prediction accuracies of estimating hourly rainfall with CART and C4.5 are 92.8% and 93.4%, correspondingly. Overall, we believe that the generated rules are useful for predicting the chance of rain and quantitatively measuring hourly rainfall.