Hybrid Deep Fixed K-Means (HDF-KMeans)

K-Means is one of the most widely used clustering algorithms due to its simplicity, scalability, and computational efficiency. However, its practical application is often hindered by several well-known limitations, such as high sensitivity to initial centroid selection, inconsistency across different runs, and suboptimal performance when dealing with high-dimensional or non-linearly separable data. This study introduces a hybrid clustering algorithm named Hybrid Deep Fixed K-Means (HDF-KMeans) to address these issues. This approach combines the advantages of two state-of-the-art techniques: Deep K-Means++ and Fixed Centered K-Means. Deep K-Means++ leverages deep learning-based feature extraction to transform raw data into more meaningful representations while employing advanced centroid initialization to enhance clustering accuracy and adaptability to complex data structures. Complementarily, Centered K-Means improve the stability of clustering results by locking certain centroids based on domain knowledge or adaptive strategies, effectively reducing variability and convergence inconsistency. Integrating these two methods results in a robust hybrid model that delivers improved accuracy and consistency in clustering performance. The proposed HDF-KMeans algorithm is evaluated using five benchmark medical datasets: Breast Cancer, COVID-19, Diabetes, Heart Disease, and Thyroid. Performance is assessed using standard classification metrics: Accuracy, Precision, Recall, and F1-Score. The results show that HDF-KMeans outperforms traditional K-Means, K-Means++, and K-Means-SMOTE algorithms across all datasets, excelling in overall accuracy and F1 Score. While some trade-offs are observed in specific precision or recall metrics, the model maintains a solid balance, demonstrating reliability. This study highlights HDF-KMeans as a promising and effective solution for complex clustering tasks, particularly in high-stakes domains like healthcare and biomedical analysis.