Optimizing Prediction Accuracy in High-Dimensional Data: Comparative Analysis of Feature Selection Methods with Naive Bayes Algorithm

This high-dimensional data is becoming increasingly common in various sectors, such as the social sciences, biology, medicine, and finance. It is defined by datasets that include many characteristics or dimensions. This study explores the idea of high-dimensional data, the difficulties it presents, and how it affects prediction results. Developing strategies to extract useful information from high-dimensional data is crucial due to its many issues, including the curse of dimensionality, increased processing complexity, and poor interpretability. This article presents research that employs feature selection techniques to solve problems related to high-dimensional data. The crucial process of feature selection is determining which features are most relevant and keeping them while eliminating those that are unnecessary or redundant. This method seeks to increase prediction accuracy, decrease overfitting, and improve model interpretability by lowering the dimensionality of the data. The present study examines three distinct feature selection methods, including Filter (SU & RELIEF), Wrapper (GENETIC & SFS), and Hybrid (combining filter & wrapper), to choose relevant features from high-dimensional data. We use a real-world high-dimensional customer dataset from UCI to illustrate how well our suggested feature selection methods work with the Naive Bayes machine learning algorithm. We demonstrate how feature selection strategies, both with and without feature selection, lead to improved prediction outcomes in high-dimensional data settings via a number of experiments and evaluations. The results show that using feature selection enhances the accuracy of the predictions. In contrast to filter and wrapper techniques, hybrid FS selects the best feature set from the three FS models. Researchers and practitioners working with high-dimensional data may make better decisions using these insights, eventually improving the prediction models' quality and applicability.