International Journal of Information Studies Mohamed El Ghourabi, Amira Dridi, Fedya Telmoudi 3 2 2011 Financial crises are perceived as shocking events, several researchers concentrated on the identification of stressed and stable periods in order to take strategic decisions on time. In this paper, a new hybrid approach is proposed to deal with the prediction of the Value at Risk (VaR). Based on financial variables from bank’s balance sheets as input data , this approach integrate Rough Set Theory (RST), Gaussian Case Based Reasoning- clustering (GCBR-Clustering), Real valued Genetic Algorithm (RGA) with Support Vector Machines (SVM) in order to classify stressed and non stressed periods and therefore determine the VaR . The RST-GCBR Clustering-RGASVM combination is justified by a high accuracy rate which reaches 96.551% in cluster 1and 100% in cluster 2.