Integrating AI for infectious disease prediction: A hybrid ANN-XGBoost model for leishmaniasis in Pakistan

Abstract

Addressing leishmaniasis infection remains a substantial challenge in KP-Pakistan due to the increased infection prevalence. Understanding its spreading tool offerings is a major challenge. We essentially design effective approaches to pinpoint its emergence and implement upgraded management strategies. This study aims to assess the prevalence of leishmaniasis infection in KP's four high-endemic districts (Bannu, Karark, Lakki Marwat, and Dera Ismail Khan) and estimate the potential future incidence. We executed a broad logical evaluation on data obtained from the pertinent district health departments of KP, using a novel hybrid ANN-XGBoost approach. We assessed its performance by equating it with frequently used models for infectious disease forecasting over time, comprising the ARIMA, LSTM, ANN, and XGBoost. We evaluated the model's precision using manifold indicators: MAE, RMSE, and MAPE. We developed the models using Python 3.11 software. The results show that the hybrid model outperformed all other models, attaining an MAE score of 82.2, an RMSE of 111.6, and a MAPE of 10.6, validating superior forecast accuracy. According to our proposed model, about 1,017 new leishmaniasis cases are expected per month by December 2025. These findings provide valuable insights for disease monitoring and intervention in KP. Advanced machine learning techniques can help policymakers improve resource mapping and come up with targeted management measures to stop the spread of leishmaniasis. Subsequent research should include other environmental and socio-economic variables influencing illness spread to improve predictive models.