Abstract — To remain competitive in the dynamic world of software development, organizations must optimize the usage of their limited resources to deliver quality products on time and within budget. This requires prevention of fault introduction and quick discovery and repair of residual faults. In this paper a new approach for predicting and classification of faults in object-oriented software systems is introduced. In particular, faults due to the use of inheritance and polymorphism are considered as they account for significant portion of faults in object-oriented systems. The proposed fault prediction model is based on supervised learning using Multilayer Perceptron Neural Network. The results of fault prediction are analyzed in terms of classification correctness and some other standard criteria. Based on the results of classification, faulty classes are further analyzed and classified according to the particular type of fault. The classification model is based on clustering using Radial-Basis Function Neural Network. It is concluded, that the proposed model provides high accuracy in discrimination between faulty and fault-free classes.

While it has long been recognized that software process improvement requires measuring both the process and its performance, experience has also shown that few universal metrics exist. The most effective measurement tools are specialized to some aspect of the task or domain being measured. This condition motivated our initial academic work with Shyam Chidamber in objectoriented (OO) metrics, which we released in 1991 1 and enhanced and formalized as a suite of six metrics in 1994 (hereafter, the CK metrics). 2 Our 1998 study explored the relationships between metrics and various outcomes of