Abstract

Dealing with an increasing complexity of software, and thus, software projects, managing those projects become more and more difficult. While monitoring a software project is a necessary task to identify potential failures or quality loss, often projects managers can only react. On the other hand, predicting the project?s progress can help project managers to avoid certain pitfalls. For this purpose, they often rely on their experience. In order to support unexperienced managers or help in new situations, a tool-based prediction using historical project data were proposed. To the best of our knowledge, all proposed prediction methods deal with metrics such as lines of code, number of defects, etc. As project managers are usually more interested in metrics such as CPI and SPI, in this thesis, we contribute with a framework that facilitates prediction for all kinds of metrics. This framework uses data mining methods to gain knowledge from past projects, and with that, predict the progress for an ongoing project. Moreover, we implemented a prediction tool that forecasts CPI and SPI values, and evaluates and visualizes the prediction results. We evaluated the prediction tool with the historical data of ITERGO. The evaluation shows promising results. Moreover, the framework is designed for adaptability and extendability. Each knowledge discovering step is encapsulated in a single component with a clear API. Moreover, the visualization as well as the data mining library are loosely-coupled and can be easily replaced. Thus, enabling the framework to be used in different problem domains.