IMPLEMENTASI SMOTE-ENN DAN BORDERLINE SMOTE TERHADAP PERFORMA LIGHTGBM PADA IMBALANCED CLASS

Abstract

Class imbalance is a significant challenge in machine learning, where unequal distribution between majority and minority classes often biases model predictions toward the majority class. This study investigates the implementation of two data balancing techniques, SMOTE-ENN (Synthetic Minority Over-sampling Technique and Edited Nearest Neighbor) and Borderline-SMOTE, to enhance the performance of the LightGBM model on the Online Shopper’s Purchase Intention dataset. The dataset exhibits an imbalanced distribution between the purchase (True) and non-purchase (False) classes, hindering the model’s ability to detect minority classes accurately. The SMOTE-ENN method combines oversampling, which creates synthetic samples for the minority class, with noise removal by eliminating misclassified samples from the majority class. On the other hand, Borderline-SMOTE generates synthetic samples near the decision boundary of the minority class, focusing on critical regions prone to misclassification. The study evaluates the LightGBM model’s performance before and after applying these techniques using evaluation metrics such as accuracy, precision, recall, and F1-score. Results demonstrate that both methods significantly improve the model’s ability to detect the minority class, with Borderline-SMOTE showing a slight advantage by generating a more representative data distribution around the decision boundary. The results indicate that both methods significantly improve the model’s ability to detect the minority class, with SMOTE-ENN achieving an accuracy of 93% and demonstrating superiority in producing a more representative data distribution compared to Borderline-SMOTE, which achieved 92% accuracy. This study confirms the effectiveness of SMOTE-ENN and Borderline-SMOTE in addressing class imbalance in machine learning applications