Serious game for Flood Prediction Using Hinge Loss Function-Based Support Vector Machine (HLF-SVM) with QOBL-BWO Feature Selection

The increasing frequency of extreme weather events worldwide has driven the need for a robust flood prediction system to anticipate and mitigate its adverse impacts.This study aims to develop a comprehensive framework by integrating the Hinge Loss Function-based Support Vector Machine (HLF-SVM) with Quasi Opposition-Based Learning and Beluga Whale Optimization (QOBL-BWO) to enhance flood prediction accuracy.The dataset was collected from the Meteorological, Climatological, and Geophysical Agency or Badan Meteorologi, Klimatologi, dan Geofisika (BMKG) in Makassar city, Indonesia, encompassing variables such as rainfall, water level, water discharge, tides, temperature, and humidity.Data preprocessing steps included data cleaning, Min-Max normalization, and the implementation of the Synthetic Minority Oversampling Technique (SMOTE) to address class imbalance.The proposed QOBL-BWO methodology was applied for feature selection, eliminating less relevant factors while simultaneously optimizing SVM parameters.The optimized HLF-SVM model was then integrated into a serious game environment to provide interactive flood scenario simulations for educational and training purposes.Test results showed 99% accuracy, 98.8% precision, 99.6% recall, and 99.2% F1-score, surpassing the performance of conventional machine learning models such as LR, LDA, KNN, NB, SVM, and AdaBoost.These findings underscore the potential of the combined HLF-SVM and QOBL-BWO approach in improving flood prediction accuracy while simultaneously promoting risk awareness and more effective decision-making.Furthermore, this framework also simplifies computational complexity in the flood prediction process.