The Large Scale Artificial Intelligence Applications - An Analysis of Al-Supported Estimation of OS Software Projects

摘要

We present the practical aspects of large scale Al-based solutions, by analysing an application of Artificial Intelligence for estimation of Open Source projects being hosted on the leading platform for Open Source - Sourceforge.net. We start by introducing the steps of data extraction task, that transformed tens of tables and hundreds of fields, originally designed to be used by web-based project collaboration system, into four datasets-dimensions important to the project management i.e skills, time, costs and effectiveness. Later, we present the structure and results of experiments, that were performed using various algorithms i.e. decision trees (C4.5, RandomTree and CART), Neural Networks and Bayesian Belief Networks. Later, we describe how metaclassification algorithms improved the prediction quality and influenced the generalization ability or prediction accuracy. In the final part we evaluate the deployed algorithms from practical point of view, presenting their characteristic beyond purely scientific perspective.