Data Mining Driven Learning Apprentice System For Medical Billing Compliance
Umair Abdullah · 2012
This research practically demonstrates how to use data mining technology to supply knowledge to the rule based system. It lays down a framework for utilization of data mining concepts to provide a sustained supply of knowledge to a rule based system at runtime. A novel ‗record couple‘ based production rule mining algorithm has been proposed to extract production rules from large datasets. Production rules thus extracted are added to the knowledge base of a rule based system. Conventional programming techniques are not efficient for performing knowledge oriented data consistency checks, especially in the domain involving large body of regularly updating knowledge.Rule based systems‘, an established technique of Artificial Intelligence (AI), is well suited to perform data consistency checks on large datasets and perform corrective measures accordingly.A problem that limits the use of rule based systems is their implementation environment, which differs from data storage environment of the business world.Rule based systems and other AI related systems are normally developed in their specific environments such as Prolog, LISP, OPS-5 etc., while data of the business world is stored in relational environment. This research proves that rule based programming is more efficient than conventional programming techniques.Further, to increase the utility of rule based systems, relational database environment has been used for implementation. A Structured Query Language (SQL) based representation of production rules has been proposed. Rule based engine has been designed, developed, and evaluated in relational database environment. Moreover, the problem of knowledge acquisition has been persistent in the area of rule based systems.Manual knowledge editing techniques have been used so far to add knowledge to rule based systems. Although machine learning techniques have been used to overcome knowledge acquisition bottlenecks, yet limited success has been reported so far. In recent years, data mining has emerged as a useful technique to solve knowledge acquisition problems. Most of the data mining work produces very useful information for business executives and decision makers; however it leaves to the choice of decision makers either to use it or to disregard that information which limits the utilization of data mining technology considerably.Additionally, the area of production rule mining‘ is not adequately explored by the data mining research community and often it is mixed up with ‗association rule mining‘. A system integrated with machine learning module is termed as Learning Apprentice System‘. Configuration of the system emerging as result of this research is termed as the ‗architecture of a learning apprentice system‘ as it involves a rule based system module‘ integrated with data mining based learning module‘. ‗Production rule mining algorithm‘ and feeding the extracted production rules to a rule based system designed in relational database environment‘ are novel contributions of this research. Medical billing compliance has been used as testing domain for concepts and ideas developed during this research.The rule based system is populated with the knowledge of medical claim processing rules using a knowledge editor.The proposed production rule mining algorithm has been utilized to mine medical claim processing rules which are then applied by the system to scrub medical claims i.e. detecting errors and performing minor, legitimate, and corrective actions. Performance of the proposed system has been proved to be efficient and has problem-solving ability as compared to the systems based on conventional programming.Conventional programming based systems are good enough to perform small and simple checks, however where complex and knowledge oriented checks are involved, the techniques proposed have proven to be much better. Evaluation of data mining driven rule based learning apprentice system proved that a lot of time were saved by prompt identification of knowledge oriented data consistency errors from the medical billing data. Although the current system has been tested in medical billing compliance domain, it can be applied to many real life problem domains which involve large numbers of knowledge oriented data consistency checks, such as credit card processing system, loan approval system, examination system etc.