Genetic, Tree, Stochastic Search (GTSS)-Based Approach for Feature Selection and Neural Network Optimization

In the realm of machine learning, the quality of the dataset plays a pivotal role in achieving accurate predictions. However, high-dimensional data often presents a challenge, laden with irrelevant features, outliers, and noise that can detrimentally affect model performance and computational efficiency. To address this pressing concern, this thesis introduces a novel approach: a genetic algorithm, tree search, and stochastic search (GTSS)-based feature selection technique. This technique sustains a population of high- performing agents. The proposed strategy leads to improved classification performance and a reduced number of features compared to conventional methods, as demonstrated by outperforming state-of-the-art feature selection techniques across multiple benchmark datasets. Building upon this foundation, the thesis further introduces a GTSS-based technique for neural network optimization. This approach seamlessly integrates genetic algorithm, tree search algorithm and stochastic algorithm into the process of weight and bias adjustment within neural networks. Leveraging the principles of natural selection and genetic operations, the proposed methodology maintains a population of high-performing network configurations, facilitating the exploration of optimal parameter configurations.

The primary objective of this approach is to enhance the accuracy of neural networks in classification tasks. Experimental evaluations conducted on various datasets showcase the efficacy of the proposed technique, consistently achieving higher classification accuracies compared to conventional optimization methods. These results underscore the potential of GTSS-based optimization techniques to significantly enhance neural network training methodologies, offering promising avenues for further research and application in real- world scenarios.