DBRF: Random Forest Optimization Algorithm Based on DBSCAN

• Published in: INTERNATIONAL JOURNAL OF ADVANCED COMPUTER SCIENCE AND APPLICATIONS
• Main Authors: Zhuo, Wang; Ahmad, Azlin
• Format: Article
• Language: English
• Published: SCIENCE & INFORMATION SAI ORGANIZATION LTD 2024
• Subjects: Computer Science
• Online Access: https://www-webofscience-com.uitm.idm.oclc.org/wos/woscc/full-record/WOS:001344145700001
• ISSN: 2158-107X; 2156-5570

correlation and redundancy of features will directly affect the quality of randomly selected features, weakening the convergence of random forests (RF) and reducing the performance of random forest models. This paper introduces an improved random forest algorithm-A Random Forest Algorithm Based on DBSCAN (DBRF). The algorithm utilizes the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm to improve the feature extraction process, to extract a more efficient feature set. The algorithm first uses DBSCAN to group all features based on their relevance and then selects features from each group in proportion to construct a feature subset for each decision tree, repeating this process until the random forest is built. The algorithm ensures the diversity of features in the random forest while eliminating the correlation and redundancy among features to some extent, thereby improving the quality of random feature selection. In the experimental verification, the classification prediction results of CART, RF, and DBRF, three different classifiers, were compared through ten-fold cross-validation on six different-sized datasets using accuracy, precision, recall, F1, and running time as validation indicators. Through experimental verification, it was found that DBRF algorithm outperformed RF, and the prediction performance was improved, especially in terms of time complexity. This algorithm is suitable for various fields and can effectively improve the classification prediction performance at a lower complexity level.