Optimisasi Metode Klasifikasi KNN, SVM, dan SVM Kernel pada Prediksi Potabilitas Air dengan Pendekatan Hyperparameter

Abstract

his research focuses on optimizing the K-Nearest Neighbor (KNN), Support Vector Machine (SVM), and SVM Kernel classification methods using a hyperparameter approach to improve the accuracy of water potability prediction. The water potability dataset containing 3276 water samples with 10 features is used. The research methods include exploratory data analysis (EDA), data preprocessing (handling missing values, standardization, dimensionality reduction, and feature selection), modeling (KNN, SVM, SVM Kernel), hyperparameter optimization using GridSearchCV, and model evaluation (accuracy, precision, recall, F1-score, ROC AUC). The results show that the SVM model with an optimized RBF kernel has the best performance, but the overall model accuracy is still not optimal. Further research is suggested to address class imbalance, select more relevant features, engineer features, and use ensemble techniques. With further development, it is expected that a more accurate and reliable model for predicting water potability can be produced, thereby contributing to improving public health and well-being.