Performance evaluation of deep neural networks for forecasting time‐series with multiple structural breaks and high volatility

The problem of automatic and accurate forecasting time-series data has always been an interesting challenge for the machine learning community. A majority real-world problems have non-stationary characteristics that make understanding trend seasonality difficult. applicability popular deep neural networks (DNNs) as function approximators TSF is studied. following DNN models are evaluated: Multi-layer Perceptron (MLP), Convolutional Neural Network (CNN), RNN with Long Short-Term Memory (LSTM-RNN) Gated-Recurrent Unit (GRU-RNN). These methods evaluated over 10 Indian financial stocks data. Further, performance evaluation these DNNs carried out in multiple independent runs two settings forecasting: (1) single-step forecasting, (2) multi-step forecasting. show convincing (one-day ahead forecast). For (multiple days forecast), different forecast periods evaluated. demonstrates long adverse effect on performance.