Stacked-ensemble forecasting of cutaneous leishmaniasis in M'Sila Algeria with epidemiology and seasonality over three decades.

Algeria hosts the largest focus of cutaneous leishmaniasis (CL) in North Africa and is among the most affected countries worldwide. This study aimed to comprehensively characterize demographic and clinical aspects, assess seasonal patterns, and develop a robust model to predict the monthly number of CL cases in M'Sila province, one of the consistently highest CL-burdened provinces of Algeria. Surveillance data (1995-2024, detailed demographics 2013-2024) sourced from M'Sila's Department of Public Health, were analysed. Seasonality was assessed via X-13 decomposition and seasonal amplitude (SA). A stacked ensemble (Kernel Ridge Regression, Random Forest, and the eXtreme Gradient Boosting) with linear meta-learner was used to predict monthly cases of CL. The performance comparison to individual learners was conducted using the RMSE/MAE/MAPE metrics. Over 30 years, CL has shown a cyclical endemo-epidemic dynamics with winter picks; 68% of cases occurred between November and January, SA=3.6, and multi-year cycles were supported by harmonic regression. From 2013-2024, 32,365 laboratory confirmed cases were reported (male-to-female ratio of 1.4) with 87.4% aged 0-30 years and 84.2% presenting a single lesion. The stacked ensemble outperformed individual learners on held-out months. The findings emphasize that CL in M'Sila exhibits pronounced seasonality and multi-year cycles. Stacked ensemble forecasting improved accuracy and can inform targeted pre-season interventions and surge planning.