Understanding customer purchasing behavior is essential for predicting customer loyalty, which directly impacts a company's long-term success. This research aims to determine the effect of chi-square and information gain feature selection in optimizing customer loyalty classification performance, compared to pure kNN. Using a public customer purchasing behavior dataset from Kaggle, containing 10,000 data, 12 attributes with loyalty_status as the label (Gold, Regular, Silver). Evaluating performance by accuracy, kappa, classification error, recall, precision, and RMSE. The highest accuracy 91.99% was obtained by kNN k=3 with information gain, kappa 0.844, precision 95.44%, recall 86.30%, with the lowest classification error 8.01% and the second lowest RMSE 0.245, after kNN k=3 with chi-square. Results show that feature selection has a positive impact on classification, increasing accuracy and reducing errors, with the combination of the kNN k=3 method and information gain proving successful in obtaining high accuracy in classifying customer loyalty.
