Application of dimensionality reduction and clustering techniques for the analysis of Carrion's disease cases in the period 2000–2024

Evangelista Gamarra M.; Chancan Labajos J.A.; Figueroa Rosas P.E.; Alvarez Manrique R.E. · Frontiers in Artificial Intelligence (2026) · DOI: 10.3389/frai.2026.1883357

Resumen

The heterogeneous geographic distribution and the complex dynamics of Carrion's disease challenge conventional epidemiological surveillance in Peru. To address this, this study applied unsupervised machine learning to 43, 534 national records (2000–2024). Following a rigorous data cleaning process—which resolved duplicate records, missing information, and outliers using Tukey's interquartile range (IQR)—the dimensionality reduction approaches MCA and FAMD coupled with the K-Means algorithm were evaluated. The calibration of the Silhouette, Davies–Bouldin, and Calinski–Harabasz indices determined that the combination of FAMD and K-Means provided the best clustering quality, identifying two clearly differentiated epidemiological profiles. An ablation control experiment demonstrated that, even after excluding the ICD-10 diagnostic coding, the geometric structure of the clusters remained highly stable, indicating that the temporal, geographic, and demographic variables contain sufficient information to preserve the clustering structure. This internal consistency was indicated through bootstrap resampling simulations. In conclusion, the coupling of FAMD and K-Means establishes a stable and reproducible framework for advanced exploratory epidemiology, constituting a valuable complementary tool to support public health surveillance and guide strategic decision-making in public health. Copyright