Explainable transformers in financial forecasting

Vasanthi Govindaraj 1, *, Humashankar Vellathur Jaganathan 2 and Prakash P 3

1 National General (An Allstate Company) Dallas, Texas, United States.
2 CGI, Atlanta, Georgia, United States.
3 School of Computer Science and Engineering, Vellore Institute of Technology, Chennai, India.
 
Review Article
World Journal of Advanced Research and Reviews, 2023, 20(02), 1434–1441
Article DOI: 10.30574/wjarr.2023.20.2.1956
 
Publication history: 
Received on 28 August 2023; revised on 21 November 2023; accepted on 24 November 2023
 
Abstract: 
This study presents a novel transformer-based model specifically designed for financial forecasting, integrating explainability mechanisms such as SHAP (SHapley Additive exPlanations) values and attention visualizations to enhance interpretability. Unlike previous models, which often compromise between accuracy and transparency, our approach balances predictive accuracy with interpretability, allowing stakeholders to gain deeper insights into the factors driving market changes. By revealing critical market influences through feature importance and attention maps, this model provides both robustness and transparency, catering to the needs of high-stakes financial environments.
 
Keywords: 
Transformers; Financial Forecasting; Explainability; Stock Prediction; XAI; Time Series Analysis
 
Full text article in PDF: 
Share this