FORECASTING THE OCCURRENCE OF DIARRHEA DISEASE USING FACEBOOK PROPHET, STATISTICS AND MACHINE LEARNING MODELS

Diarrhea disease remains a significant public health concern among children under 5 years of age, necessitating reliable forecasting tools to support timely intervention and healthcare planning. This study applies a combination of classical time-series models and supervised machine learning techniques to model and forecast monthly diarrhea cases using data obtained from childhood diseases records in Rivers State Hospital Management Board, in collaboration with agencies responsible for childhood disease surveillance. The  dataset spans January 2015 to December 2024 and represents population-wide hospital-reported cases, collected under a rigorous sampling framework to ensure data credibility. All analyses were conducted using Python. Exploratory time-series analysis revealed persistent temporal dependence and seasonal fluctuations. Forecasting models implemented include FB Prophet with linear and logistic trend specifications [11, 12], ARIMA(1, 0, 1) [25], SARIMA(1, 0, 1) (1, 0, 1) [3, 26], Random Forest [13], Artificial Neural Network (ANN) [14], and a surrogate Long Short-Term Memory (LSTM) model [26]. The Prophet linear model assumes constant growth with additive monthly and yearly seasonality, while the logistic variant incorporates saturation effects through an externally defined carrying capacity. ARIMA and SARIMA models captured short-term persistence and seasonal shocks, whereas machine learning models were configured to learn nonlinear and temporal patterns in the data. Model performance was evaluated using Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). Results indicate that the ARIMA model achieved the best overall performance (RMSE = 124.17, MAPE = 21.49%), closely followed by the ANN (RMSE = 124.29, MAPE = 24.55%). The SARIMA model showed slightly inferior accuracy, suggesting limited seasonal dominance in the series. Among machine learning models, ANN outperformed Random Forest and LSTM, while the Prophet logistic model performed better than its linear counterpart but lagged behind ARIMA and ANN. The findings demonstrate that diarrhea incidence in Rivers State exhibits both short-term persistence and moderate nonlinear behavior. Linear time-series models and neural networks provide superior predictive accuracy, underscoring their suitability for disease surveillance and public health decision-making. Integrating statistical and machine learning approaches offers a robust framework for forecasting childhood diarrhea and optimizing healthcare resource allocation.