Representative Case-Based Retrieval to Support Case-Based Reasoning for Prediction

Abstract

Case-based reasoning (CBR) is used to resolve a new problem by searching in past similar situations. It is widely used for prediction. However, CBR induces many shortcomings, particularly in its retrieval phase. Many methods and techniques were provided, but have influenced differently the effectiveness of the deduced results. Also, hybridizing different methods emerged to get a better information retrieval and this reasoning manner becomes a ubiquitous issue. The present study has as an objective lending support to CBR to enable enhanced retrieving valid prediction. For this purpose, the study proposes a methodology based on hybridizing data mining with CBR. Thereby, a data mining model is used and a reduced search space solution before processing a CBR's prediction retrieval is proposed. To assess the approach, a clustering was applied to a car safety data set to generate a reduced case base. Then CBR uses it for predicting the car safety. The results show a precision over 80% and an accuracy over 82%, which are well over the classical CBR and indicate the relevance of the approach.