Weighted Deep Learning Approach For Better Forecasting

There are different kinds of time series analysis and forecasting techniques can be found in the literature. The prediction of unknown future values based on known historical data is one of the goals to be achieved. Here, another approach by combining well-known Deep Learning methods with Weighted Moving Average method is introduced. The Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) from the Recurrent Neural Networks family are utilized in this study. We also compare the prediction results of the proposed approach, namely weighted LSTM (w-LSTM) and weighted GRU (w-GRU), with the original implementation of LSTM and GRU. Different scenarios using real-world import values dataset are developed in the experimentation phase. It was found that the proposed approach could get lower Root Mean Square Error, Mean Absolute Error, and Mean Absolute Percentage Error at 1143.242, 999.028, and 0.155 respectively than the original Deep Learning methods.