Comparison of Multi-layer Perceptron and Support Vector Machine Methods on Rainfall Data with Optimal Parameter Tuning

Abstract

This study describes the search for optimal hyperparameter values in rainfall data in 49 cities in Australia, consisting of 145,460 records with 22 features. The process eliminates missed values and selects 16 numeric type features as input features and one feature (Rain Tomorrow) as output feature. It is processed using the Multi-Layer Perceptron (MLP) and Support Vector Machine (SVM) methods based on Three Best Accuration (3BestAcc) and Best Three Nearest Neighbors (3BestNN). The results showed that the SVM kernel linear method gave an average accuracy value of 0.85586 and was better than the MLP method with an accuracy of 0.854.