Performance Evaluation of the Random Forest Algorithm in Classifying Breast Tumors Based on Histological Image Feature

Abstract

      This study aims to evaluate the efficiency of a classification approach based on medical image analysis for diagnosing breast tumors and distinguishing between benign and malignant patterns. The analysis relied on two sets of digital histological images, comprising 10,000 images in the first set and 2,400 images in the second, after subjecting them to a series of preprocessing steps that included image enhancement, dimension standardization, and the use of texture-analysis techniques to extract features that describe the tumor’s micro-structural characteristics. A classification model was then constructed using the Random Forest method, and its performance was assessed through a comprehensive set of statistical indicators. The findings demonstrate a high capability of the model to differentiate between the two classes, achieving elevated accuracy levels and displaying stable performance despite the variation in sample size. The results further reveal that the Random Forest algorithm exhibits statistical robustness and consistent performance, while larger samples contribute to stronger discriminatory power and improved stability of the evaluation metrics. Overall, the study underscores the significant role of preprocessing and texture-feature extraction in enhancing diagnostic quality, supporting the potential use of this approach in medical applications concerned with tumor characterization.