Comparing Effectiveness of Statistical Versus Deep Learning for Time Series Forecasting

One of the most popular academic subjects is the forecasting of time series. Machine learning, or ML, approaches have been proposed as substitutes to statistical ones in academic literature for forecasting time series. However, information on their respective effectiveness when it comes to both precision and computational demands is scarce. Machine learning approaches are increasingly being used to handle these predicted challenges, instead of traditional statistical methods. However, there is a limitation of comparative data for advanced models developed in the past few years such as DeepAR and Prophet. We have also utilized models such as LSTM, GRU, RNN, ARIMA. In this research, we addressed this problem and compared six different models for the Jena Climate dataset and the Appliances Energy forecast dataset.