AI-DRIVEN DATA QUALITY MANAGEMENT AND ANOMALY DETECTION IN LARGE-SCALE E-COMMERCE DATABASES: A COMPREHENSIVE ANALYSIS WITH FOCUS ON UZBEKISTAN'S DIGITAL MARKETPLACE ECOSYSTEM

Аннотация

Contemporary e-commerce environments face unprecedented challenges in maintaining data integrity across vast transnational databases. This investigation examines artificial intelligence applications for automated data quality control and anomaly identification within high-volume digital commerce platforms. Our empirical study, conducted across multiple e-commerce ecosystems including Uzbekistan's rapidly expanding digital market, processed 62.7 million transactions over eight months. Machine learning implementations achieved 91.8% precision in detecting data inconsistencies while reducing manual oversight requirements by 72%. Ensemble- based detection systems demonstrated 38% superior performance compared to conventional rule-based approaches, particularly in identifying fraudulent patterns and database anomalies. The Uzbekistan market analysis revealed unique data challenges including multi-currency processing, diverse payment integration, and multilingual content management, providing valuable insights for emerging digital economies. Keywords: Data Quality Assurance, Machine Learning Detection, Digital Commerce, Database Optimization, Uzbekistan E-commerce 1. Introduction Digital commerce platforms worldwide experience exponential transaction volume growth, creating complex data management challenges that traditional approaches struggle to address effectively. Uzbekistan's e-commerce sector exemplifies these challenges perfectly, with over 50 marketplaces generating $300 million annually, projected to reach $1 billion by 2027. Uzum, the leading platform, serves nearly a third of the nation's population monthly, processing 14 million orders in the first nine months of 2024 with daily GMV capacity of $4.5 million. Data quality issues manifest through duplicate customer records, inconsistent product categorization, pricing anomalies, and fraudulent transaction patterns. With Uzbekistan's eCommerce market projected to grow by 11.12% annually through 2029, reaching $2.6 billion, addressing these challenges becomes critical for sustainable digital commerce growth. Traditional rule-based systems prove inadequate when https://scientific-jl.com/wsrj 2 4 1 Volume-40_Issue-1_June-2025 World scientific research journal confronting the scale and complexity of modern e-commerce operations, necessitating sophisticated AI- driven solutions. 2. Research Methodology This investigation employed a comprehensive mixed-methods approach, combining quantitative performance analysis with qualitative assessment of implementation challenges across diverse e- commerce environments over eight months (January-August 2024). 2.1 Data Sources and Collection Framework The research dataset comprised 62.7 million transactions, 4.2 million customer profiles, and 1.8 million product entries from three platforms: Platform Alpha (International Electronics): 22.4 million transactions, $287 average order value, 45 countries Platform Beta (Regional Fashion): 18.6 million transactions, $73 average order value, Central Asia focus Uzum Market (Uzbekistan): 21.7 million transactions, $52 average order value, 600,000+ SKUs, 9,000+ merchants, multilingual support (Uzbek, Russian, English) 2.2 Algorithm Implementation Four machine learning approaches were implemented: Random Forest Classifier, Gradient Boosting (XGBoost), Support Vector Machines, and Isolation Forest for unsupervised detection. An ensemble methodology combining weighted voting and stacking approaches was developed for optimal performance across diverse data types and cultural contexts. 3. Results and Analysis 3.1 Data Quality Issue Distribution Analysis revealed distinct patterns across platforms, with Uzbekistan showing unique characteristics: Issue Category Platform Platform Beta Uzum Total % Alpha Market Missing Data Fields 31,200 42,150 38,920 25.4% Format Inconsistencies 22,480 28,340 41,680 20.9% Duplicate Records 18,420 24,680 31,240 16.8% Product Classification 14,720 21,450 19,280 12.5% Address Standardization 12,680 16,820 24,360 12.1% Currency Conversion Errors 3,240 5,670 18,440 6.2% Payment Validation 4,180 6,240 15,680 5.9% https://scientific-jl.com/wsrj 2 4 2 Volume-40_Issue-1_June-2025 World scientific research journal exhibited significantly higher currency conversion errors and payment validation issues, reflecting complexity in handling UZS-USD transactions and integrating multiple local payment systems (Click, Payme, Uzcard, Humo). 3.2 Machine Learning Performance Performance evaluation revealed ensemble approaches yielding optimal results: Algorithm Precision Recall F1-Score AUC-ROC Ensemble Model 0.918 0.921 0.920 0.952 Gradient Boosting 0.916 0.902 0.909 0.938 Random Forest 0.893 0.876 0.884 0.921 Support Vector Machine 0.871 0.859 0.865 0.904 Isolation Forest 0.858 0.834 0.846 0.887 The ensemble model's superior performance was particularly pronounced in the Uzbekistan data set, attributed to its capability to simultaneously handle diverse data types, multiple lan