Comparative Analysis of Time Series Models for the Prediction of Conjunctivitis Disease

Abstract

Conjunctivitis is one of the most common disease in Hong Kong. The health department and government of Hong Kong are taking several actions to prevent the occurrence of this disease. But unfortunately, every week many cases of conjunctivitis are reported in Hong Kong. Therefore, the advance prediction of future occurrences of conjunctivitis cases can help the government to take pre-action to curb that. Time series forecasting techniques can be utilized in this scenario to predict the future occurrences of the same. In this manuscript, various state of the art time series models viz. Auto Arima, Exponential Smoothing (ETS), Neural Network (NN), and Random Forest (RF) are applied and a comparative analysis between all is made. The error metrics used for evaluation are: Mean Error (ME), Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and Auto Correlation Function (ACF). The results show that Random forest performs best by decreasing all the error metrics by a substantial amount in comparison to auto arima, NN, ETS respectively.