• Across 1991–1997, research coalesced around automatic rule discovery and rule mining, unifying text categorization, relational data, and association patterns into compact, transferable models learned from large data stores [2], [3], [9], [5], [6], [1], [19], [14].

• Database-centric mining emerges as a core paradigm, treating data stores (databases, data warehouses) as enabling platforms for discovery, performance-focused mining, and attribute-oriented learning across relational DBs and warehousing contexts [1], [8], [9], [16], [15].

• Methods to handle continuous or mixed-mode data began with discretization to enable inductive learning, converting attributes into discrete forms while preserving predictive power, enabling rule induction on real-valued domains [4], complemented by data-driven rule discovery on heterogeneous data [14].

• Pattern discovery extended beyond single rules to structured representations: higher-order neural networks extract production rules; sequential patterns enable probabilistic induction; and clustering perspectives explore non-metric structure, signaling richer knowledge extraction [6], [18], [11], [17].

• Efficiency, quality, and robustness motivate pruning and selective evaluation of rule sets: pruning for compact rules, adjustable accuracy, and noise-tolerant learning to extract reliable models from noisy, large-scale data [5], [19], [3], [2], [10].