A Comparative Study For Imbalanced Data Techniques Of Classification Algorithms

Abstract

One of the main challenges in data processing using machine learning is the imbalanced data distribution, where minority classes are often underrepresented, leading to biased predictions in classification algorithms such as K-Nearest Neighbors (KNN), Naive Bayes, and Support Vector Machine (SVM). This study aims to address this issue by applying Random Undersampling (RUS), Synthetic Minority Oversampling Technique (SMOTE), and hybrid approaches such as SMOTETomek. Using the NHANES dataset, this study evaluates the effectiveness of these methods in reducing bias and improving classification performance. The hybrid sampling technique performed the best, increasing sensitivity to minority classes, resulting in more balanced predictions. Models tested using metrics such as accuracy, precision, recall, and F1-score showed that SVM achieved the highest accuracy of 98.8% after hyperparameter tuning. This study also emphasizes the importance of hyperparameter optimization, including parameters such as C and gamma for SVM, k values ​​for KNN, and smoothing factors for Gaussian Naive Bayes, to improve model reliability. These findings emphasize the importance of effective data preprocessing techniques and model optimization in dealing with imbalanced datasets. Implementing these approaches will ensure more accurate data analysis, as well as provide valuable insights for decision-making and policies aimed at improving imbalanced case.