Performance Analysis of Deep Learning Techniques for Time Series Forecasting

Abstract

Time series data forecasting is a crucial topic in economics, business, and finance. New methods are being created to evaluate and predict time series data as a result of recent improvements in computing power and more significantly the development of sophisticated machine learning algorithms and methodologies, such as deep learning. This work boons a deep learning-based time series forecasting method and a comparative study among three models that are ARIMA, LSTM and FB-Prophet by using the current knowledge as time series then draws out the key elements of prior data to forecast the values of an upcoming time sequence. In this work, we have taken an electric production dataset, from which 70% of data used as training and 30% of data used as testing the methods. From the experimental result it is found that ARIMA model outperforms the other two model in forecasting this timeseries data.