Rainfall Prediction in Tengger, Indonesia Using Hybrid Tsukamoto FIS and Genetic Algorithm Method

Abstract

Countries with a tropical climate, such as Indonesia, are highly dependent on rainfall prediction for many sectors, such as agriculture, aviation, and shipping. Rainfall has now become increasingly unpredictable due to climate change and this phenomenon also affects Indonesia. Therefore, a robust approach is required for more accurate rainfall prediction. The Tsukamoto Fuzzy Inference System (FIS) is one of the algorithms that can be used for prediction problems, but if its membership functions are not specified properly, the prediction error is still high. To improve the results, the boundaries of the membership functions can be adjusted automatically by using a genetic algorithm. The proposed genetic algorithm employs two selection processes. The first one uses the Roulette wheel method to select parents, while the second one uses the elitism method to select chromosomes for the next generation. Based on this approach, a rainfall prediction experiment was conducted for Tengger, Indonesia using historical rainfall data for ten-year periods. The proposed method generated root mean square errors (RMSE) of 6.78 and 6.63 for the areas of Tosari and Tutur respectively. These results are better compared with the results using Tsukamoto FIS and the Generalized Space Time Autoregressive (GSTAR) model from previous studies.

Keywords

• genetic algorithm
• hybrid
• prediction
• rainfall
• roulette wheel
• Tsukamoto FIS.