Enhancing Prediction Accuracy using Ensemble Deep Learning Methods on Time Series Data

Time series prediction, which has numerous applications in several fields like electricity consumption management, business, healthcare, and so on, is addressed as one of the core tasks of data analytics. Traditional methods are often not able to capture the complicated sequences inherent in real-data. Nowadays, deep learning methods are investigated to forecast the time series pattern. However, the expected accuracy of the prediction is not obtained in many cases. In this paper, we propose an ensemble method that unifies the recurrent neural network (RNN) and convolutional neural network (CNN) models into one framework, allowing for improved predictive performance. In the proposed ensemble method, several models, like the long short-term memory (LSTM) recurrent neural network, gated recurrent unit (GRU), and temporal convolutional network (TCN) are used as weaker models and extreme gradient boosting (XGBoost) is used to combine an ensemble from the baseline models. In the ensemble method, a more accurate feature set is generated, put through dense layers, and the combined features are used for further predictions. In the experimental results, it is observed that the proposed ensemble method outperforms several efficient single models and hybrid models on real-time time series datasets.