Machine Learning and Decision Support in The Exploitation of Mdx Queries: Case of Banking Products and Services

Abstrak

The increasing complexity of analytical workloads in banking systems challenges traditional query scheduling mechanisms, particularly in OLAP environments where MDX queries exhibit heterogeneous computational costs and business criticality. This study proposes a semantic-aware scheduling framework based on incremental machine learning to dynamically prioritize MDX queries in real time. The approach models query prioritization as a classification problem, integrating both technical features and business-driven criticality.

A Hoeffding Tree algorithm is employed to enable continuous learning from streaming query data without requiring retraining. The model is evaluated using a simulated dataset of 10,000 MDX queries reflecting realistic banking scenarios, including risk monitoring and regulatory reporting. Experimental results show that the proposed approach achieves a classification accuracy of 94.1% and significantly reduces processing latency for high-priority queries, with improvements reaching 42.4% compared to FIFO scheduling. The inference overhead remains negligible, ensuring compatibility with real-time system constraints.

These findings demonstrate the effectiveness of integrating incremental learning into query scheduling and highlight the potential of semantic-driven optimization in decision support systems. The study contributes to bridging the gap between learned database systems and business-aware query management.