OUTLIER DETECTION IN DATA WAREHOUSES TO IMPROVE DESCRIPTIVE AND DIAGNOSTIC ANALYTICS

This paper reviews the literature on outlier detection(OD) technologies to improve descriptive and diagnostic analytics in data warehouses. This will ensure higher-quality and more reliable data, increasing decision-making and operational efficiency. The major research objectives are a systematic review of existing OD techniques; identification and discussion of the key challenges and limitations in applying the OD methods to data warehouse environments and synthesis of methodologies for integrating OD with descriptive and diagnostic analytics. The study collects data from both traditional and AI-based literature review tools. Traditional review tools are Google Scholar, Research Gate, and IEEE Xplore. AI-based review tools are Semantic Scholar, Research Rabbit, and SciSpace. To present the insights, the study objectively selected 57 papers that were published between 2010 and 2024 were considered. The literature review here elaborates on the OD in a data warehouse which uses different data warehousing techniques and data analytics to enhance the quality and reliability of data to correct. This systematic review goes from the evaluation of statistical-based to distance-based, density-based, clustering-based, learning-based, and ensemble-based. The OD methods of unsupervised learning have been found to outperform those of supervised learning in special settings that are massive and heterogeneous in information like data warehouses. Isolation Forest, Local Outlier Factor (LOF), One-Class SVM, and Autoencoders are identified as highly accurate and efficient at detecting anomalies. Moreover, hybrid models combining several OD methods have been demonstrated to perform better than individual techniques. The results may be useful in offering important new insights and practical guidelines for developing more effective.