Metric-based neural network classification tool for analyzing large-scale software

Abstract

The neural network described performs classification of software metrics. It is a three-layer, error back-propagation network. Using historical data, the neural network learns the relationship between certain metrics and a particular classification. The neural network selects the classification which best fits the input metrics. The capability of neural networks to classify nonlinearly separable problem spaces gives them an advantage over tree-based and linear network-based classification methods. When applied to actual software metrics, the neural network correctly classified 100% of the data presented.<>