tsta

Home / دومین کنفرانس ملی عصر انفجار تکنولوژی؛ هوش مصنوعی، تحولی در صنعت، تجارت و زنجیره تامین و دومین کنفرانس ملی علم داده در کاربردهای مهندسی

بهبود راهکار انتخاب ویژگی ترکیبی با ارزیابی یکپارچه‌ روابط خطی و غیرخطی ویژگی‌ها

Authors :

سید مجتبی سیف¹

1- دانشگاه آزاد اسلامی واحد صفاشهر

Keywords :

Feature selection،Pearson correlation coefficient،Normal mutual information،Clustering،Ensembling feature selection indicator

Abstract :

Feature selection is a crucial task in data science that significantly influences the performance and efficiency of machine learning algorithms. Many existing approaches depend on a single linear or nonlinear correlation measure, which often fails to capture the intricate dependencies among features, thereby limiting model effectiveness. To overcome this limitation, we propose a novel composite metric—Absolute Pearson Correlation and Normalized Mutual Information (APCNMI)—which jointly captures linear (via the Pearson correlation coefficient) and nonlinear (via normalized mutual information) relationships among features. Based on this metric, we develop a three-stage hybrid feature selection algorithm. The first stage applies an APCNMI-based filter to eliminate irrelevant features. The second stage uses APCNMI-driven feature clustering to group redundant features, significantly narrowing the search space. In the final stage, Particle Swarm Optimization (PSO), a population-based metaheuristic algorithm, is employed to select a compact and high-quality feature subset from the reduced space. Extensive experiments on six real-world datasets demonstrate the superiority of the proposed method. Results show that our algorithm consistently selects smaller yet more informative feature subsets and yields better classification performance compared to methods relying on a single correlation measure.