Big Data-Driven Time Series Forecasting for Financial Market Prediction: Deep Learning Models

Journal of Artificial Intelligence and Big Data | Vol 2, Issue 1

Table 1. Summary of the related work on Time SeriesForecasting in the financial market

Ref	Methodology	Dataset	Performance	Limitations & Future Work
Ma et al. (2019)	Hybrid model combining ANN and SVR with dynamic adjustment bias	China Stock Exchange data (June 8, 2015 – May 26, 2016)	79% accuracy rate	Future work could involve more diverse market conditions and cross-market validation
Liu and Liu, (2018)	GRU with movement trend-based preprocessing (two-step: trend extraction + discretization)	Predicting the pattern of stock index movement	Accuracy improved from 33% to 68%	Needs validation across different indices and longer time frames
Tsang, Deng and Xie, (2018)	Deep LSTM-based time-series prediction	To predict the next day's closing price, six global market indicators	Annual profitability up to 200%	Risk-adjusted returns and robustness under volatile conditions not explored
Raimundo and Okamoto, (2018)	SVR-Wavelet hybrid model using DWT for input enhancement	FOREX market time series data	Improved financial series prediction accuracy	Scalability to other financial domains and high-frequency data not addressed
Althelaya, El-Alfy and Mohammed, (2018)	Deep RNNs (LSTM and GRU, uni- & bi-directional, stacked) with multivariate input	S&P500 historical index data	Superior to shallow networks in short/long-term forecasting	Needs improved interpretability and computational efficiency for real-time systems
Beyaz et al. 2018	Machine learning with clustering to account for market mood states	Forecasting selected company stock prices with 126-day horizon	Mood-based forecasting improved accuracy in 47% cases	Further study needed on integrating sentiment analysis with fundamental and technical data
Bakhach, Tsang and Jalalian, (2017)	Directional Change (DC) framework for trend prediction	Forex market (EUR/CHF, GBP/CHF, USD/JPY)	Accuracy often exceeded 80%	Limited to one independent variable; expansion to multivariate models is needed