SUMMARY
ANFIS is a combination of 2 methods, namely fuzzy and artificial neural networks. From the combination of these 2 methods, ANFIS has the advantage of predicting time series data. However, the use of the ANFIS method also has its drawbacks. The drawback is that it can decrease the accuracy obtained. The decrease is because the data used is widely error-value and does not vary. In addition, the data used does not go through the preprocessing process first. Data collection comes from the BMKG website. The data will later be preprocessed to overcome the imbalance class. The results of training and testing with the process of optimizing the ANFIS method by normalizing and cleaning erorr data on rainfall data there are few similarities in both training and testing. For performance measurement, prediction accuracy uses RMSE for both 3:2 and 4:1 data ratios – getting 0.0728 and 0.0539, respectively. Based on the results of the application of the ANFIS method and normalization in the rainfall dataset of Sleman regency with parameters of the number of membership functions, input membership functions, learning rate, data ratios of 3:2 and 4:1 shows that the ANFIS method with data that has been carried out the normalization process and cleaning of error data can be an alternative method to predict rainfall levels with time series data.