Anomaly detection plays a vital role for e-commerce businesses due to the enormous volume of numeric data they handle.   Price data is a prominent example of such data, and detecting anomalies in numeric data is challenging due to potential errors and outliers, which can disrupt sales calculations and result in financial losses. Anomaly detection in e-commerce aims to identify the outliers within a given dataset automatically. This research study explores unsupervised and supervised machine learning approaches for detecting price anomalies in publicly available e-commerce datasets. This research proposed an anomaly detection framework that involves feature selection using the Z-score technique, followed by multiple corresponding analyses (MCA) of the same selected features. Machine learning techniques, including Isolation forest and DBSCAN, are employed to detect price anomalies. A threshold is then established for a voting scheme (Voting ZID) to enhance the accuracy of the anomaly detection technique.  Different machine learning classifiers are evaluated after labelling the anomalies (anomalous and non-anomalous), including Gaussian Naïve Bayes, Random Forest, LDA, SVM and the proposed Gradient Boosting Machine technique. The algorithms are optimized by tuning different parameters.  The proposed technique attains high performance in terms of precision 0.9713 % and accuracy 0.9751 % - the highest on the benchmark.

Keywords: Machine learning, Anomaly Detection, Algorithm, and E-commerce.