SOL-AUTOCLUST: A SMART ONLINE-LEARNING AUTOMATED CLUSTERING FRAMEWORK

The automation of machine learning has predominantly focused on supervised tasks, leaving unsupervised clustering, a critical component of exploratory data analysis, significantly underdeveloped by existing Auto-ML frameworks. Current approaches often limit their scope to dataset characteristics, neglecting the crucial influence of algorithmic suitability (e.g., robustness to outliers) and user-defined requirements (e.g., interpretability needs). This oversight leads to suboptimal clustering outcomes, particularly when dealing with complex, high-dimensional, or noisy data. To address these limitations, this research introduces SOL-Auto-Clust, a novel end-to-end automated clustering framework that makes a key contribution by holistically integrating three fundamental dimensions: inherent data characteristics, intrinsic algorithmic traits, and explicit user-defined objectives. By employing a meta-feature architecture, SOL-Auto-Clust dynamically generates customized clustering pipelines, addressing both data intricacies and real-world application requirements. Extensive evaluation across diverse datasets highlights the framework's ability to simplify clustering processes and produce reliable, insightful outcomes, marking a significant step towards human-aligned Auto-ML for unsupervised learning.