CLASSIFICATION AND CLUSTERING PROBLEMS IN MACHINE LEARNING ASSISTED BY METAHEURISTIC ALGORITHMS

Abstract

The field of machine learning has witnessed remarkable advancements in recent years, driven by the ever-increasing availability of data and the development of various algorithms. In this context, the importance of feature selection and clustering in data analysis is crucial. Feature selection plays a vital role in overcoming the curse of dimensionality and identifying the most relevant and non-redundant features for classification tasks. On the other hand, clustering, particularly k-means, is widely used in data analysis, but determining the optimal number of clusters and initializing cluster centroids remain challenging tasks. This thesis explores the utilization of Metaheuristic Algorithms (MAs) to address these challenges. MAs are a class of optimization techniques that are particularly well-suited for solving complex problems where traditional optimization methods may fall short. Specifically, this thesis focuses on four MAs: Particle Swarm Optimization (PSO), Grey Wolf Optimizer (GWO), Artificial Hummingbird Algorithm (AHA), and Arithmetic Optimization Algorithm (AOA). These MAs are applied to enhance the efficiency and accuracy of feature selection in classification tasks and centroid initialization in clustering problems. Feature selection is a critical preprocessing step in machine learning, aimed at selecting a subset of relevant features from a high-dimensional feature space. The curse of dimensionality poses a significant challenge in this regard. The thesis demonstrates how MAs, with their powerful local and global searching capabilities, efficiently find optimal feature subsets. By doing so, they contribute to the construction of robust machine learning models that can better generalize from the data. The practical use of MAs in feature selection is illustrated through real-world scenarios. For instance, a Levy flight Grey Wolf Optimizer (LW-GWO) is employed for i breast cancer diagnosis. This application showcases how the LW-GWO algorithm efficiently selects the most informative features from a vast dataset, leading to improved accuracy in diagnosis. Clustering is a fundamental unsupervised learning technique used to group similar data points together. However, it requires determining the optimal number of clusters and initializing cluster centroids effectively. The thesis delves into the integration of MAs, specifically AHA and AOA, for these purposes. The clustering process becomes more efficient, resulting in improved clustering results with the help optimization capabilities of MAs.