A Fast Scalable Calssifier Tightly Integrated with RDBMS

In this paper, we report our successin building efficient scalable classifiers by exploring the capabilities ofmodern relational database management systems (RDBMS). In addition to highclassification accuracy, the unique features of the approach includeits high training speed, linear scalability, and simplicity inimplementation. More importantly, the major computation required in theapproach can be implemented using standard functions provided by themodern relational DBMS. Besides, with the effective rule pruningstrategy, the algorithm proposed in this paper can produce a compactset of classification rules. The results of experiments conducted forperformance evaluation and analysis are presented.