ANALYSIS OF FILTER-BASED FEATURE SELECTION METHODS USING TOP FEATURE SUBSETS AND MULTIPLE EVALUATION METRICS

Abstract

Feature selection is an important step in improving the performance and efficiency of machine learning models, especially when handling high-dimensional datasets. The presence of irrelevant and redundant features can reduce model accuracy and increase computational cost. In this study, several widely used filter-based feature selection techniques are examined, including Chi-Square, Analysis of Variance (ANOVA), Mutual Information, and correlation-based methods. The main objective is to analyze how these techniques perform when selecting different numbers of top-ranked features.To ensure a fair evaluation, multiple datasets with different characteristics are used. For each dataset, feature subsets are created by selecting the top 20, 40, 60, and 80 features based on ranking scores. These subsets are then used to train machine learning models, and their performance is evaluated using metrics such as accuracy, precision, F1-score, and Root Mean Square Error (RMSE). This approach helps in understanding how the number of selected features influences model performance. The results show that feature selection has a significant impact on model outcomes. In many cases, better performance is achieved with a moderate number of features rather than using all available features. It is also observed that the effectiveness of each method varies depending on the dataset. Overall, this study provides useful insights for selecting appropriate feature selection techniques in practical machine learning applications.