Machine learning and Linear method Which Methods Provide Better Forecasts

Machine learning methods to forecast are increasingly applying in different disciplines and are used in wide range of applications. Predicting financial time series is achallenging area of forecasting in finance as well as machine learning. The paper aims to compare a machine learning method (nonlinear model) and a linear forecasting method in gold price forecasting. To do so, GMDH-type (Group Method of Data Handling) neural network, as nonlinear method, which uses an evolutionary method and ARIMA forecasting model as a linear method are employed. Our Results show that GMDH type neural network makes a better forecast in comparison with ARIMA model, based on MAPE and MPE criteria.