MODELLING HOSPITAL ADMISSION RATES IN SÃO PAULO, BRAZIL LEE-CARTER MODEL VS. NEURAL NETWORKS

In Brazil, hospital admissions represent almost 50% of the total claims cost of health insurance companies while they only represent 1% of the total medical procedures. Therefore, modeling hospital admissions is extremely useful for health insurers to assess their claim costs over time and actuaries should be capable to include that information in their analyses, in order to preserve the financial sustainability of the companies. This dissertation analyses the use of the Lee-Carter model for predicting the general level of hospital admissions in the state of São Paulo, Brazil, using the traditional ARIMA model and contrasting it with the LSTM neural network. Publicly available data between the years 2008 and 2019, divided by gender, were used. The function auto.arima from the R package forecast was used to find the best ARIMA model for the data while the LSTM neural network model was searched in a combination of 20 models, varying the learning rate and decay factor. The results showed that the LSTM model and the ARIMA have similar RMSE and MAE performance.