Forecasting Malaria in Indian States: A Time Series Approach with R Shiny Integration

TL;DR

This paper develops and evaluates time series models, especially a log-transformed polynomial regression, for forecasting malaria cases in eight Indian states, and integrates these models into an interactive R Shiny tool for practical use.

Contribution

It introduces a robust forecasting approach using log-transformed polynomial regression and provides an interactive R Shiny application for malaria prediction in India.

Findings

Log-transformed polynomial regression outperformed other models in accuracy.

The models demonstrated robustness across all eight states.

The R Shiny tool enables real-time malaria forecasting and decision support.

Abstract

Malaria remains a significant public health challenge in many regions, necessitating robust predictive models to aid in its management and prevention. This study focuses on developing and evaluating time series models for forecasting malaria cases across eight Indian states: Jharkhand, Chhattisgarh, Maharashtra, Meghalaya, Mizoram, Odisha, Tripura, and Uttar Pradesh. We employed various modeling approaches, including polynomial regression with seasonal components, log-transformed polynomial regression, lagged difference models, and ARIMA models, to capture the temporal dynamics of malaria incidence. Comprehensive model fitting, residual analysis, and performance evaluation using metrics such as Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE) indicated that the log-transformed polynomial regression model consistently outperformed other models in terms of accuracy and robustness across all states. Rolling forecast validation further confirmed the superior predictive capability of the log-transformed model over time. Additionally, an interactive R Shiny tool was developed to facilitate the use of these predictive models by researchers and public health officials. This tool allows users to input data, select modeling approaches, and visualize predictions and performance metrics, providing a practical tool for real-time malaria forecasting and decision-making support. Our findings highlight the critical role of appropriate modeling techniques in malaria prediction and offer valuable resources for enhancing malaria surveillance and response efforts.