Penelitian ini bertujuan menganalisis pengaruh metode penanganan nilai kosong dan variasi jumlah fitur terhadap kinerja Random Forest pada klasifikasi kelayakan air minum. Penelitian menggunakan dataset Water Quality dengan 3.276 observasi dan 10 variabel yang terdiri atas 9 variabel prediktor dan 1 variabel target. Tahapan penelitian meliputi penanganan nilai kosong menggunakan metode drop, mean imputation, median imputation, dan KNN imputation, seleksi fitur berdasarkan feature importance Random Forest, optimasi GridSearchCV, serta evaluasi menggunakan accuracy, precision, recall, dan F1-score. Hasil penelitian menunjukkan kombinasi mean imputation dengan Top 5 Features menghasilkan performa terbaik dengan accuracy 0,6631, precision 0,5806, recall 0,4922, dan F1 score 0,5328. Berdasarkan confusion matrix, model menghasilkan 309 True Negative (TN), 91 False Positive (FP), 130 False Negative (FN), dan 126 True Positive (TP). Hasil tersebut menunjukkan bahwa Random Forest mampu melakukan klasifikasi kelayakan air minum, namun masih memiliki keterbatasan dalam mengenali kelas air layak minum yang ditunjukkan oleh nilai false negative yang masih tinggi.

This study aims to analyze the effect of missing value handling methods and feature quantity variations on the performance of Random Forest for drinking water potability classification. The study utilizes the Water Quality dataset consisting of 3,276 observations and 10 variables, including 9 predictor variables and 1 target variable. The proposed approach includes missing value handling using deletion, mean imputation, median imputation, and KNN imputation methods, feature selection based on Random Forest feature importance, GridSearchCV optimization, and evaluation using accuracy, precision, recall, and F1-score metrics. The results show that the combination of mean imputation with Top 5 Features achieved the best performance, with an accuracy of 0.6631, precision of 0.5806, recall of 0.4922, and F1-score of 0.5328. Based on the confusion matrix, the model produced 309 True Negative (TN), 91 False Positive (FP), 130 False Negative (FN), and 126 True Positive (TP). These results indicate that Random Forest is capable of classifying drinking water potability; however, the model still has limitations in identifying potable water samples, as indicated by the relatively high number of false negative predictions.