Home / Current Issue / Paper 1722225
Leakage-Aware Temporal Machine Learning for Forecasting Weekly Lassa Fever Cases: Independent-Year Validation Using NCDC Surveillance Data
Subject area: Science,Engineering and Technology · Area of research: Machine Learning
DOI: 10.64388/ire1722225
Abstract
Reliable forecasting of infectious disease incidence can strengthen surveillance and support timely public health planning. This study evaluated a leakage-aware temporal machine learning approach for forecasting weekly confirmed Lassa fever cases in Nigeria using routine surveillance data from the Nigeria Centre for Disease Control and Prevention (NCDC). The dataset comprised 313 weekly observations from 2020 to 2025. Temporal feature engineering produced 33 predictors and 261 model-ready observations after the first 52 weeks were excluded to construct annual lag features. The data were divided chronologically into training (2021–2023), validation (2024) and independent testing (2025) periods. Four regression algorithms Linear Regression, Random Forest, XGBoost and Support Vector Regression (SVR) were assessed. Optuna-Based Hyperparameter Optimization used validation RMSE as its objective. Performance on the 2025 test set was measured using MAE, RMSE, MAPE and R². SVR achieved the lowest MAE (6.3546) and RMSE (8.7611), and the highest R² (0.8148). Random Forest recorded the lowest MAPE (44.5906%). These results indicate that model performance differed by metric, although SVR provided the strongest overall balance. The study demonstrates the value of combining temporal feature construction, validation-based optimization and chronological testing when evaluating weekly Lassa fever forecasts. The 2025 results provide a held-out temporal assessment within the same national surveillance series, rather than external validation.
Keywords
chronological validation, information leakage, Lassa fever, NCDC surveillance, Optuna-Based optimization, Support Vector Regression, temporal machine learning, weekly case forecasting
How to cite this paper
@article{1722225,
author = {Wuese Paul Shir, Shammah Emmanuel Chaku, Kehinde Henry Oduwole, Yunana Kefas},
title = {Leakage-Aware Temporal Machine Learning for Forecasting Weekly Lassa Fever Cases: Independent-Year Validation Using NCDC Surveillance Data},
journal = {Iconic Research And Engineering Journals},
year = {2026},
volume = {10},
number = {1},
pages = {4262-4271},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1722225.pdf},
abstract = {Reliable forecasting of infectious disease incidence can strengthen surveillance and support timely public health planning. This study evaluated a leakage-aware temporal machine learning approach for forecasting weekly confirmed Lassa fever cases in Nigeria using routine surveillance data from the Nigeria Centre for Disease Control and Prevention (NCDC). The dataset comprised 313 weekly observations from 2020 to 2025. Temporal feature engineering produced 33 predictors and 261 model-ready observations after the first 52 weeks were excluded to construct annual lag features. The data were divided chronologically into training (2021–2023), validation (2024) and independent testing (2025) periods. Four regression algorithms Linear Regression, Random Forest, XGBoost and Support Vector Regression (SVR) were assessed. Optuna-Based Hyperparameter Optimization used validation RMSE as its objective. Performance on the 2025 test set was measured using MAE, RMSE, MAPE and R². SVR achieved the lowest MAE (6.3546) and RMSE (8.7611), and the highest R² (0.8148). Random Forest recorded the lowest MAPE (44.5906%). These results indicate that model performance differed by metric, although SVR provided the strongest overall balance. The study demonstrates the value of combining temporal feature construction, validation-based optimization and chronological testing when evaluating weekly Lassa fever forecasts. The 2025 results provide a held-out temporal assessment within the same national surveillance series, rather than external validation.},
keywords = {chronological validation, information leakage, Lassa fever, NCDC surveillance, Optuna-Based optimization, Support Vector Regression, temporal machine learning, weekly case forecasting},
month = {July},
doi = {https://doi.org/10.64388/ire1722225}
}