Currently, a group of machine vision researchers are working on the problem called Text Recognition in the Wild: dawn text recognition tasks which involve recognizing hand written or typed text from scanned images to convert them into digital form. The difficulty of this work is influenced by factors such as writing style, size, and quantity pictures used to address the problem. This article was aimed toward the offline recognition on Gurmukhi letters from text images using Multi-Layer Perceptron (MLP) Neural Network model. It is an efficient training method using a smaller number of passes and follows generalized delta learning principle. The MLP model, the system currently used in projecting isolated handwritten Gurmukhi characters to lines and subsequently identifying those lines and their relation to each character block within ASCII images has resulted in identification rate as high at 98.96%. The paper considers the use of a logistic regression (LR) model and an MLP to evaluate the PD in SMEs. The MLP aids in calculating PD values, which are then calibrated using the real portfolio default average. Results indicate that the combined MLP-LR model outperforms conventional logistic regression techniques, providing higher classification accuracy.