International Journal of Computer Science Trends and Technology (IJCST) – Volume 5 Issue 3, May – Jun 2017 ISSN: 2347-8578 www.ijcstjournal.org Page 89 Balancing the Complexity of Architecture and Generalization of Soft-Computing Model in Predicting the Properties of Composite Preforms Dr. P. Radha Associate Professor MCA Department Mepco Schlenk Engineering College, Sivakasi Tamil Nadu - India ABSTRACT The novel strategy proposed in this paper is used to reduce the complexity of architecture of soft-computing model like neural network with high accuracy in predicting the outputs. It further improves the recognition power of neural network while handling raw-data with highly non-linear, more interrelated, noisy and MAR (Missing At Random) values. The bias term was slightly modified in MRBNN (Modified Radial Basis Neural Network) to improve the generalization of over-fitting problems. The architecture of network model was balanced with the network generality in Powder metallurgy Lab for predicting the deformation and strain hardening properties of AI-Fe composite preforms. Keywords:- soft computing, Radial Basis neural network, composite preforms I. INTRODUCTION Soft computing, as opposed to conventional “hard” computing, is a technique that is tolerant of imprecision, uncertainty, partial truth and approximation. Its methods are based on the working of the human brain and it is commonly referred to as Artificial Intelligence (AI). The action of AI is similar to the human brain which is capable of arriving at valid conclusions based on incomplete and partial data obtained from prior experience. The soft computing methods are robust and low cost. The application of soft-computing tools in the material engineering was analyzed in the early research [1]. In this article, the soft-computing tool like neural networks was applied in Powder Metallurgy area to process the properties of metal powders. The soft- computing based Simulation of powder metallurgical preforms may avoid lab experiments involving dangerous materials and hence prevent risky consequences. This model is not only avoids expensive experiments but also evade handling dangerous materials that cause severe damage to environment. As per the existing models adopted in the previous research [2-5] with neural networks, the following factors were identified.  Little number of outputs is possible from more number of inputs: More inputs are used to derive one or two outputs. In this case, it is not necessary to use the soft-computing approach.The number of inputs is more than thrice amout of outputs.  No strategy, in fixing the correct combination of input features– arbitrarily & manually:  Many input features are involved in designing the soft-computing model. The relevant input feactures can be selected only by the experience of technicians in powder metallurgy Lab. Also, for each set of combination, this model gives different results. To overcome this problem, the standard approach is used to select the input features in this paper.  Applied more number of training samples: Generally, RBF needs more samples for proper training. In terms of thousands, training samples were preferred for training in previous research.  Spent more number of hidden neurons: Due to more training samples, the size of hidden layer is high in case of exact interpolation of earlier RBF models. RESEARCH ARTICLE OPEN ACCESS