A Contribution to Time Series Analysis and Forecasting Using Deep Learning Approaches

In the dynamic realm of time series analysis and forecasting, the pursuit of more precise and efficient models persists as a fundamental objective. This research contributes by presenting a comprehensive comparison involving our recent Hybrid Multivariate model (GRU-1DCNN). Our proposed model capitalizes on intricate architecture, drawing inspiration from a fusion of 1D Convolutional Neural Networks (1DCNN) and Gated Recurrent Unit (GRU) memory cells, with the primary goal of surpassing the performance achieved by previously developed custom models in this domain. Evaluation of the predictive capabilities of the models is conducted using key metrics such as Root Mean Squared Error (RMSE), R-square, Mean Absolute Error (MAE), Mean Squared Logarithmic Error (MSLE), and E-Variance. Notably, the study focusses on addressing the challenges posed by real-world data, exemplified by the volatility inherent in Market data, representing a worst-case scenario. The proposed hybrid model remains modular and exhibits applicability to diverse time series datasets, allowing for the capture of nuanced temporal patterns. This research contributes to the ongoing pursuit of advancing time series modeling techniques, offering insights into the efficacy of our hybrid model and its potential to outperform existing approaches across various domains.