Performance Analysis of Machine Learning Algorithms Using Information Theoretic Class Based Multi-correlation Dependent Feature Selection

Techniques for feature selection pick the most useful features from high-dimensional datasets to minimize dimensionality while preserving semantic information. Dimensionality reduction has a great impact on the learning accuracy and the computational complexity of techniques for deep learning and machine learning. Most feature selection techniques focused on removing irrelevant and redundant features only and ignored multi-correlation factors such complementarity and interaction between features. Motivated by this, we applied a variant of the R2CI feature selection algorithm to select the most informative features in the cervical cancer risk factor dataset and observed its impact on the accuracy of machine learning algorithms. Comparative study of machine learning algorithms’ performances is presented in this research on considering class-based feature multi-correlation and information-theoretic dependent FSA. For our study, we considered twelve different machine learning algorithms for comparison using metrics such as accuracy, precision, recall and ROC-AUC curve. Results show a significant increase in accuracy of all the machine learning algorithms that are considered.