Database Normalization via Statistical Cost Model

The patent replaces manual or rule-based schema decomposition with a machine learning model that automatically learns functional dependencies from data patterns. The ML model analyzes the denormalized table and generates normalized schemas without requiring explicit specification of dependencies by users, thereby reducing the complexity of identifying and maintaining functional dependencies while still achieving data redundancy reduction.

The system enables self-service normalization where the ML model autonomously performs schema decomposition by learning directly from the data. The model automatically identifies patterns, determines functional dependencies, and generates optimal normalized schemas without external intervention, making the process efficient and adaptive to different data types.

The patent performs preliminary actions by pre-training the ML model on various normalization patterns and functional dependency relationships. This pre-training enables the model to quickly evaluate multiple schema options and select the optimal one during actual use, thereby improving schema quality without significantly increasing selection time for new datasets.

The system changes parameters by using probabilistic scoring and confidence thresholds to evaluate schema options. The ML model generates multiple candidate schemas with associated quality metrics, and the optimal schema is selected based on these parameters, allowing for efficient comparison and selection without exhaustive evaluation of all possible schemas.