Classification of Leukemia Cancer Data using Correlation Based Feature Selection Model: A Comparative Approach

Abstract

The abundance of data obtained from microarray experiments presents challenges related to the number of variables and the presence of random fluctuations. Despite the efforts that had been made by previous researchers, emphasizing how data mining aids the implementation of models to facilitate informed prediction, gaps are evident which requires improvement over the
earlier models. Dimensionality reduction techniques, such as Correlation Based Feature Selection (CBFS), are good candidate solutions to these problems by selecting pertinent features for categorization. This research implements a model for classification of leukemia cancer using CBFS with Support Vector Machine (SVM), k-Nearest Neighbors (KNN), Decision Tree (DT), and
Ensemble classifiers. The evaluation of the performance of these machine learning models was carried out using sensitivity, specificity, precision and accuracy. The findings indicate that the CBFS+DT model outperforms the other models in terms of sensitivity (96.75%), specificity (97.18%), precision (97.56%), accuracy (96.75%), and F1 score (96.97%), while also exhibiting a decreased computational time (0.4336). This demonstrates the efficacy of CBFS in improving classification accuracy and reducing computing load. Overall, this study highlights the effectiveness of CBFS in cancer research and underscores the importance of carefully choosing the most pertinent variables to enhance classification outcomes