Explainable artificial intelligence for human-friendly explanations to predictive analytics on big data

Abstract

Nowadays, machine learning techniques have become critical for decision-making mechanisms in numerous real-life applications in areas like healthcare, justice, transportation and finance. However, recommendations made by machine learning techniques, as well as their logical reasoning behind these recommendation decisions, are often not easy to be comprehended by humans. This thesis presents an explainable artificial intelligence (XAI) solution that enhances state-of-the-art techniques to produce more understandable and practical explanations to end-users. To evaluate the practicality and usefulness of this XAI solution, a case study was conducted on a big data predictive model built based on real-life customer churn data. Results show that the presented solution successfully provides users with more friendly and useful explanations when compared to related works.