Performance comparison of different time series models for computer-aided forecasting of measles case numbers

In a globalized world where travel times have shortened, the spread of infectious diseases, particularly through contact and respiratory secretions, has become easier. The measles virus, which spreads easily through respiratory droplets, remains a deadly disease today. Despite the proven effectiveness of vaccines in largely preventing measles, the virus continues to be endemic in regions with low vaccination rates. Public health in countries receiving migrants from these regions is adversely affected. The successful computer-aided prediction of measles cases can help prevent the spread of the disease. In this study, to determine the model to be used in developing a computer-aided system, the prediction performances of time series models with different methodologies based on classical statistics (ARIMA, HW), machine learning (Greykite, Prophet, XGBoost), and deep learning (LSTM) were compared. To generalize the performance of the models, datasets containing the number of measles cases from Benin, Cameroon, and Nigeria were used. The models were trained on the first 147 months of data from the three different datasets, and their prediction accuracy was tested on the following 12 months. The performances of the models were measured using the RMSE, MAPE, MAE, and MSLE metrics. The findings show that the XGBoost model was more successful in predicting the number of measles cases in Benin (8,21 RMSE, 27,48 MAPE, 6,50 MAE, 0,08 MSLE) and Nigeria (457,39 RMSE, 84,44 MAPE, 289,22 MAE, 0,69 MSLE), while the LSTM model performed better for Cameroon (43,65 RMSE, 103,58 MAPE, 36,68 MAE, 0,67 MSLE). The developed computer-aided system forecasts the number of measles cases for the next 6 months for these countries. It was forecast that in Nigeria, the number of cases would increase approximately 43 times in the first month, decrease by more than half in the following month, remain stable for the next 4 months, and then decrease again in the final month. No sudden changes in the number of measles cases were forecast for the other countries. The superior performance of AI models compared to traditional models offers significant progress, suggesting that measuring their applicability for other infectious diseases in future studies could contribute to the field of health.