Explainable Clustering of Plasmodium Species Using Hybrid CNN-LBP and UMAP for Enhanced Classification Insight

Accurate and rapid identification of Plasmodium species from microscopic images is vital for malaria diagnosis. This study proposes a novel framework that integrates eight CNNs and LBP features with unsupervised and supervised UMAP embeddings for explainable clustering and enhanced classification insight. Hybrid CNN-LBP representations enhanced UMAP embedding compactness and separability for top-performing backbones, with variability across models as confirmed by internal clustering metrics. The framework achieved an F1-score above 0.96 and an accuracy above 0.96 using k-NN on MP-IDB as the primary dataset. However, these gains were accompanied by a decline in AUC-PR, indicating reduced probability calibration and highlighting a trade-off between accuracy and reliability. The framework maintained competitive performance on external datasets, reaching 0.86 accuracy and 0.85 F1-score. Overall, the proposed framework provides an explainable, computationally efficient, and clinically relevant approach that enhances insight into malaria species classification. © This article is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License. License details: https://creativecommons.org/licenses/by-sa/4.0/