Convolutional Neural Networks for Time Series Forecasting

Abstract

Time series forecasting occurs when you make scientific predictions based on historical time stamped data. It involves building models through historical analysis and using them to make observations and drive future Strategic decision-making. Their main specificities compared to the most common areas of machine learning are their dependence over time and their seasonal behaviors that can appear in their evolution. In the literature, statistical models are widely used for time series forecasting. However, many complex models or approaches can be very useful in some cases. Generalized Autoregressive Conditional Heteroskedasticity (GARCH), Bayesian models and ARIMA vectors (VAR) are just a few examples. There are also even time series models borrowed from deep learning. Deep learning methods do offer a lot of promise for time series forecasting, specifically the automatic learning of temporal dependence and the automatic handling of temporal structures like trends and seasonality. Multilayer Perceptrons, Convolutional Neural Networks (CNN), or Recurrent Neural Network (RNN) with LSTM or GRU cells can be successfully employed for time series forecasting issues. Through this work, we aim to develop a CNN model for time series forecasting. The proposed model will be experimented and evaluated on real-world datasets with the known metrics in this field.