An analysis on impact of feature selection in CBR performance by predicting bankruptcy

Abstract

Bankruptcy prediction is very important because it affects the organization as well as the entire nation's economy. Hence an effective bankruptcy prediction model is required. Various statistical and intelligent techniques are available to predict bankruptcy among that Case Based Reasoning (CBR) is more effective since it provides prediction along with explanation. CBR bankruptcy prediction model effectiveness depends on the feature selection technique and case retrieval algorithm used in it. There are many feature selection techniques and retrieval algorithms used in bankruptcy prediction models. In our model we use forward feature selection and backward feature elimination in order to obtain best features and K-Nearest Neighbor algorithm for case retrieval. This model also makes a comparative study on those two feature selection techniques with influencing features selected by real genetic algorithm. The results of forward feature selection yield s 82 % accuracy in bankruptcy prediction when it is compared to other feature selection techniques.