Research On Approach To Software Classification Based On Decision Tree

Software classification avails to analyze and evaluate software so that to plan the process properly and organize human resource effectively in the process of software development. The off-the-shelf approach classifies software only by outer attributes. The result is imprecise and it cannot classify software in fine grain. Thus, to find a new approach to software classification is required.Aimed at classifying software in fine grain, an approach to software classification through data mining is proposed. The method of decision tree is chosen to build the classification model based on the traits of data collected during software testing with the request of classifying software in fine grain. Among the decision tree algorithms, SLIQ is chosen to generate the classifier of software for its good performance on dealing with numeric attributes. As the MDL-based pruning algorithm is not credible enough when pruning branches after the tree-building step, pre-pruning category is used instead. K-fold cross-validation is used to verify the performance of the classification models.Different testing editions of DM are selected to build classification system on, including steps of data collecting and preprocessing, tree building, model verifying, rule explaining and analysis. The experimental results indicate that the rules conform with the truth. Attributes that affect the final label of software mostly are picked out. Useful knowledge can be retrieved from the rules to guide the following development processes, which proves the practical value of software classification approach based on decision tree.