Deep Learning Algorithm Forecasting the Unemployment Rates in the Central European Countries

Abstract

The aim of this paper is to forecast the monthly unemployment rate’s time series using deep learning algorithms. Based on data from five Central European countries, we tested the forecasting performance of the ‘conventional’ Box–Jenkins methodology in comparison with three deep learning models: the CNN (Convolutional Neural Network), the MLP (Multilayer Perceptron) and the random forest algorithm. The MAPE, MAE, RRMSE, and MSE error tests were used for testing the forecasting results. In our results, the ARIMA model was outperformed by one of the deep learning algorithms in all cases. The medium-term predictions suggest that in the Central European area, unemployment will remain relatively high in the future.