An integrated approach for combination reduction detection using a hybrid optimization technique on gene expression Data

One of the most crucial challenges within high-dimensional data analysis in the field of machine learning revolves around the process of Feature Selection (FS). Due to the large number of features in micro array datasets, only less number of them are important for analysis. Extraction of relevant features is therefore essential. Multiple feature selection methods are available, with filter techniques standing out in this domain due to their versatility with various learning algorithms, substantial reduction in optimization time, and overall improvement in model performance. Additionally, the way a filter technique is applied depends on both dataset characteristics and the chosen machine learning model. As a result, a combined approach involving feature reduction (CFR) followed by a widely recognized optimization technique like Particle Swarm Optimization (PSO) is being considered. Four filtering strategies are taken into account, and sixteen (4*4) pipeline combinations are produced in the previous work where researcher get the one feature reduction is works better in all the model specified. In this research its main objective is to implement the same better feature sequence in the different classifier with four microarray datasets followed by one optimization technique. As here classification approaches include Naïve Bayes (NB), and Random Forest (RF). The pipeline includes several filter techniques, namely Correlation-Based Feature Selection (CBFS), Relief Feature Selection (RFS), Information Gain (InG), and Chi-Square Test (CST). The datasets and classifiers have a big impact on how effectively CFR performs. The Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) approach is then used to rank all classifiers among all, over the top feature reduction sequence.