Akademik Veri Yönetim Sistemi
10TH INTERNATIONAL CONFERENCE ON ADVANCES IN STATISTICS, Budapest, Macaristan, 19 Nisan - 21 Kasım 2024, ss.49
Big data can be analyzed, and important information can be extracted by using data mining and machine learning methods. The biggest challenge in information extraction is the presence of many unnecessary variables that reduce the performance of machine learning
methods and provide disadvantages and complexity in terms of calculation time. Feature selection methods are part of data preprocessing to overcome this problem. It aims to improve the performance of machine learning methods by selecting distinctive features that best represent the data. Feature selection methods can be classified into two main approaches: filter and wrapper. In wrapper methods, feature selection is a binary NP-hard problem. One approach to solving NP-hard problems is meta-heuristic algorithms. However, the metaheuristic algorithms cannot be used directly to solve binary problems without modifications. For this reason, this study modifies the Chernobyl Disaster Optimizer (CDO) to binary CDO (BCDO) and uses this algorithm as a wrapper-based feature selection technique. In addition, well-known UCI datasets are tested and compared with other feature selection methods to
demonstrate and verify the performance of the BCDO algorithm. Various evaluation measures are used to evaluate the performance of the BCDO algorithm, including accuracy, the number of selected features, fitness values, sensitivity, specificity, and convergence curves.