Advancing Classification Algorithm Selection with Ensemble Meta-Learning- A Data-Driven Approach

Selecting the optimal classification algorithm for diverse datasets remains a significant challenge in data mining. This study introduces a robust Ensemble Meta-Learning (EML) framework that automates the selection process by leveraging the diversity of meta-features. Our approach dynamically adjusts the number of recommended algorithms based on dataset character- istics, demonstrating substantial improvement over traditional methods. Through empirical evaluations on 183 datasets and 20 classification algorithms, EML outperforms established single- link and ML-KNN-based methods in terms of recommendation accuracy and offers more precise algorithm selection. This paper underscores the potential of ensemble meta-learning in enhancing algorithmic recommendations, paving the way for future innovations in meta-learning applications.