Anomaly Analysis on E-Commerce Seller Performance Using an Intersection Multi-Method Anomaly Detection Approach

Abstrak

The growth of e-commerce in Indonesia has led to an increasing number of sellers with increasingly diverse performance characteristics. Variations in seller performance have the potential to generate patterns that deviate from the majority of the data, making anomaly identification essential for supporting data-driven evaluation and decision-making. This study aims to detect anomalies in seller performance and cluster these anomalies based on the identified characteristics. The study employs the Knowledge Discovery in Databases (KDD) framework, which consists the stages of data selection, preprocessing, transformation, and data analysis. Anomaly detection is performed using a multi-method approach that combines Isolation Forest (IF), Histogram-Based Outlier Score (HBOS), One-Class Support Vector Machine (OCSVM), Copula-Based Outlier Detection (COPOD), and Empirical Cumulative Distribution Outlier Detection (ECOD) through an intersection-based approach. Subsequently, the K-Means algorithm is used to cluster the anomalous data based on the similarity of their characteristics. The results show that the proposed approach successfully identifies sellers with characteristics that different from the majority of the data. The clustering process produces two main clusters with a Silhouette Score of 0.7754, a Davies-Bouldin Index of 0.3702, and a Calinski-Harabasz Index of 65.8701. These clusters represent positive and negative anomalies based on the seller performance characteristics.

Identifier

• DOI: https://doi.org/10.67561/jaids.v2i2.3
• OpenAlex: https://openalex.org/W7170160914