Comparative stability analysis of mixed clustering algorithms for Malaysian dengue epidemiology using topological descriptors.

Abstract

Dengue fever remains a significant public health challenge in Malaysia, with high case numbers reported annually. Effective control and mitigation strategies require robust analytical tools to understand transmission dynamics and guide interventions. This study utilized topological data analysis (TDA) to extract structural features from epidemiological dengue time series using Euler characteristic curve. In contrast to traditional clustering methods that rely on direct application to dataset, TDA-based approach encodes qualitative topological information which is robust to noise and effectively capture underlying transmission dynamics. A comparative stability analysis is conducted by introducing controlled perturbations (noise) to the input data and the performance are accessed using multiple external validation metrics. Taking the k-medoids clustering algorithm with 30 % noise as an example, the TDA-based clustering approach (using the Euler characteristic) demonstrated significantly greater robustness and stability across all evaluation metrics. The Adjusted Rand Index (ARI) improved by 252 %, while the Normalized Mutual Information (NMI) increased by 122.5 %. The Fowlkes-Mallows Index (FMI) rose by 67.3 %. Additional improvements are seen in V-Measure (122.5 %), Homogeneity (80 %), and Completeness (169.7 %), highlighting the superior performance of the TDA-based approach under noisy conditions. These results demonstrate the noise resistance of the TDA-based clustering method, highlighting its enhanced ability to preserve meaningful cluster structures under perturbation compared to traditional approaches. This improved robustness facilitates deeper insights into disease transmission patterns and supports more effective data-driven public health analysis.