Optimizing Regression Algorithm Performance for Weak Rainfall Dataset Prediction Via Ensemble Machine Learning Models

• Main Authors: Sudarno, Prabowo Wahyu, Ashari, Ahamd, Riasetiawan, Mardhani
• Format: Other NonPeerReviewed
• Language: English
• Published: Journal of Theoretical and Applied Information Technology 2022
• Subjects: Analysis of Algorithms and Complexity
• Online Access: https://repository.ugm.ac.id/284220/1/104.%20Optimizing%20Regression.pdf; https://repository.ugm.ac.id/284220/; https://www.researchgate.net/publication/374840735
• Institution: Universitas Gadjah Mada

A flood is a natural disaster that cannot be stopped, but preventive measures can be taken to deal with it. The factors that cause flooding can be predicted using machine learning, one of which is by predicting rainfall. But in reality, rainfall data has many shortcomings, such as missing values and the appearance of outliers that can affect model performance. Therefore, we propose an ensemble stacking method to deal with this problem. The performance value of the Multilayer Perceptron algorithm without Stacking is 10.128 for MSE and1.5696 for MAE. The performance value of the XGBoost algorithm without stacking is 9.2548 for MSE and 1.4427 for MAE. While the performance value of combining the Multilayer Perceptron and XGBoost algorithm with Stacking resulted in an MSE value of 9.2377 and an MAE value of 1.4396. The results show that the ensemble method with stacking can be a solution to improve algorithm performance on weak datasets to predict rainfall value. The novelty of this paper is as follows: machine learning ensembles can handle the weak rainfall dataset to give a better result