A Probabilistic Model For Automobile Diagnosis System: Combining Bayesian Estimator And Expert Knowledge

Recent trends in Artificial Intelligence based softwares have a strong link up with learning. Probabilistic graphical models have been used over the years for solving problems under uncertainty. In this paper, an automobile diagnosis system is proposed to predict the root reason for a faulty part inside a car engine. The system combines Conditional Probabilistic Distributions (CPDs) from the expert as well as those learnt from the user using a Bayesian estimator. In this regard, a learning function is incorporated to combine the CPDs in terms of weighted mean. These combined CPDs are then modeled by a Bayesian Network that is traversed to return a probabilistic solution according to the symptoms given by the user. The Variable elimination algorithm is used for inference. In this regard, several variable ordering heuristics have been evaluated and compared in terms of time efficiency.