Predicting Issue Handling Process using Case Attributes and Categorical Variable Encoding Techniques

Software development process deals with the increasing complexity, needs for enhancement, and the reduction of bugs. The software project managers have a responsibility to maintain the software quality, e.g., by establishing Service Level Agreement (SLA) to the software development. However, there are difficulties to monitor SLA and to predict the software quality, due to a lot of issues should be handled. This work proposed prediction techniques for handling issues in the software development by using historical data from software repositories. There are two types of predictions in this work, namely 1) prediction of the remaining duration, and 2) prediction of the next activities. Event logs extracted from historical data in the software repositories exploited case and event attributes as predictors. Furthermore, the categorical variable encoding technique used in the preprocessing data phase. Some categorical variable encoding techniques also proposed in this study. The results showed that the use of case attributes as predictors could improve performance by 8.57% in the next activity prediction and 1.9% in the remaining duration prediction. Of the 4 (four) categorical variable encoding techniques used, one-hot encoding and sum encoding techniques can provide the best remaining duration prediction with a MAE of 19.72 and one-hot encoding technique can provide the best next activity prediction with an accuracy of 65.38%. All of these methods are widely evaluated using datasets from the Google Chromium Project. Further work including utilization of other attributes outside case and event attributes, e.g., component attributes, as predictor variables.