An Optimized Machine Learning Based Rule Engine Architecture

Abstract

Rules are segments of knowledge which are generally conveyed as, “when some conditions are evaluated as true, then perform some tasks". A rule engine is basically an advanced software system responsible for rules evaluation and execution. While it is easy to add rules, problems arise when their numbers tend to explode exponentially over time due to new business scenario needs. It eventually becomes more complex when an enterprise information system, having configuration model powered by a declarative rule engine, needs to be maintained. Performance significantly degrades when there are thousands of rules, since the engine must figure out every time which rule should be fired when large number of facts arrive for processing, thus rapidly choking up the system. Objective of this paper is to use machine learning techniques to optimize declarative business rules system when the number of rules increases creating performance degradation and complexity issues.