K-Means Post-Training Optimizer

We present a novel approach to optimize artificial intelligence (AI) models using unsupervised learning, specifically utilizing the k-means clustering algorithm with varying cluster parameters. The proposed method aims to enhance the performance of the initial model by iteratively evaluating it after adding new features to the original dataset. The novelty is the use of k-means model outputs as a new feature. By tracking the k-means model outputs we find the best tuning parameters to make the best testing evaluation results and feed the outputs back into the original feature set. This iterative process continues until the accuracy of the metamodel reaches a convergence point. The methodology is designed to uncover latent patterns within the data through the exploration of different cluster configurations, fostering a dynamic feature extraction process. We demonstrate the effectiveness of our approach on the Diabetes Dataset, and Wisconsin Breast Cancer Detection Dataset showcasing improvements in model accuracy as a result of the iterative integration of evaluation data. Our experimental results highlight the adaptability and robustness of the proposed method. Additionally, we discuss the implications of our findings for various applications and provide insights into potential avenues for future research in the realm of unsupervised learning for post-training model optimization.