EXPLAINABLE AI TECHNIQUES IN DATA SCIENCE A SYSTEMATIC LITERATURE REVIEW
Keywords:
Explainable Artificial Intelligence, XAI, Machine Learning, Data Science, Interpretability, Systematic Literature Review, PRISMA, Deep Learning, CNN, LIME, SHAPE.Abstract
The increasing adoption of Artificial Intelligence (AI) in data science has enhanced predictive capabilities in various fields, including healthcare, cybersecurity, finance, education, e-commerce, and industrial analytics. Many current AI systems, however, are black-boxes, which lack transparency, interpretability and trust [1]. To address these challenges, Explainable Artificial Intelligence (XAI) has become an important research area that offers explanations of model predictions that are understandable. This systematic literature review (SLR) aims to collate the research works related to XAI techniques in data science from 2020 to 2026. Based on the PRISMA methodology, relevant studies were identified by conducting structured searches in Scopus, IEEE Xplore, ACM Digital Library and Web of Science, which resulted in 21 studies being selected. The review classifies XAI techniques into model-agnostic, intrinsic, visualization-based, attention-based, and counterfactual approaches and discusses various application areas, evaluation metrics, tools, challenges, and directions for future research. Results reveal that the most prevalent approaches in the existing literature are SHAP, LIME, Grad-CAM, and attention-based methods, and that there are limitations with respect to the standardized metrics for evaluation, performance-interpretability trade-offs, scalability, and understanding by humans. The review results offer a systematic and reproducible synthesis and suggest future research directions for trusted and explainable AI systems.
