Machine Learning to Detect Fraud: Mixing Traditional Indian Wisdom with Modern Technology

With the exponential growth of e-commerce, return fraud has become a critical concern for online retailers, impacting profits and customer trust. Modern computational techniques, especially machine learning (ML), offer advanced capabilities to detect complex fraud patterns. This paper explores the use of ML algorithms—Logistic Regression and Gradient Boosting—for return fraud detection using a dataset of 101 e-commerce transactions. Uniquely, this study draws conceptual inspiration from the Indian Traditional, particularly from Ayurveda and Jyotish Shastra, which are related with multi-feature analysis and temporal pattern recognition, respectively. Ayurveda’s principle of diagnosing health imbalances through multiple bodily and behavioural features is mapped to ML’s feature-based fraud analysis. Jyotish’s emphasis on time and cyclic behaviour informs the study’s focus on time-series patterns in fraud behaviour. By comparing logistic regression with gradient boosting on the dataset, we demonstrate that Gradient Boosting outperforms in detecting non-linear and time-sensitive fraudulent behaviours. This research bridges the relation between traditional system and new computational science.