The protein structure has always been under significant exploration, as this is vitally responsible for the basic functionality. Understanding the formation of these structures has gradually been called as `the protein folding problem'. The solution for this problem is basically concerned with the ultimate aim to attain the native state. Broadly the methods to predict the ultimate goal, are categorized as- the template -based (homology modeling, threading/fold recognition) and template -free (ab initio) methods, which are discussed later in detail. In this paper, we have followed the ab initio methodology, to develop an ARTNN based approach to cluster the stable folds, out of the search space. The approach implemented through either of the two models (SF-ART or FC-ART model) could help enhance the performance of obtaining the three dimensional native state by providing more productive pathway, using the crude ab initio parameters, emphasizing quality performance with reduced execution time. This could thus be considered as a productive ab initio- clustering approach.