CONMI_MFS: A New Filter Feature Selection Algorithm

Feature selection is crucial in machine learning and data mining. Existing feature selection algorithms often use single evaluation metrics for assessing relationships between variables or have high computational complexity. Taking into account the intricacies of how features interact with each other, we have developed a novel Improved feature selection evaluation function known as CONMIS. This function combines the Pearson correlation coefficient, which measures linear relationships, with maximized mutual information, which captures nonlinear dependencies, to provide a thorough characterization of the inter dependencies between features and the target class variables. Building on this, we introduce the CONMI_MFS algorithm designed to identify an optimal feature subset that is both highly relevant to the class variables and exhibits minimal redundancy. Through comparative experiments with four alternative methods across ten datasets, our results indicate that CONMI_MFS can achieve a maximum reduction rate of 67.63%. This underscores the competitive edge of the CONMI_MFS algorithm in terms of performance.