EFFECTIVENESS OF DEEP LEARNING MULTISTEP FORECASTING MODEL IN TIME SERIES FORECASTING

Abstract

Forecasting time series is an essential part of many applications. However, because of the intricate statistical makeup of time series data, modelling financial time series can be difficult. To find the key characteristics of the data and random variation, time series analysis is a helpful technique. The ability of machine learning and deep learning algorithms to address real-world issues has attracted attention in recent years. Despite the importance of time series forecasting, order dependence between observations makes it a difficult endeavour. This paper's major goal is to suggest a paradigm for comparison that addresses these restrictions. Time series forecasting is more challenging than simple classification and regression issues because of the order dependence between observations. The suggested methodology carries out comprehensive experiments to assess the performances of different statistics, Machine Learning, and Deep Learning methodologies available in the literature for prediction in order to address this issue. The analysis is further expanded to include hybrid models and ensembles to determine whether merging individual models increases prediction accuracy. Due to their ability to uncover intricate hidden patterns in time series datasets that traditional statistical techniques are unable to, the review suggests that Deep Learning approaches are the best predictors. Deep learning is thus a useful technique for time series forecasting. The paper provides insights into this area by examining the most recent developments in deep learning as they are applied to time series forecasting.