Bug-fix time prediction with machine learning

Abstract

Bug-fix time is an essential part of any bug report. Accurate bug-fix time predictions help in many project management related issues. This thesis examines, whether it is possibly to find out a suitable machine learning model for making accurate bug fixing time predictions based on bug report data. Project is implemented for ATR Soft Oy and the needed data is fetched from ATR Soft work hour registration system and open-source bug tracking system. Thesis begins with fundamentals of machine learning, while also covering popularly used machine learning algorithms. Similarly, natural language processing basic concepts are introduced. Literature review is conducted for finding important machine learning characteristics from previous bug-fix time estimation papers. These found characteristics are then utilized in the machine learning process for creating a machine learning model, which predicts bug fixing times. Suitable categories for predictions are formed and data preprocessing is performed. After the initial results, machine learning model is tuned for final results of the project. Based on the evaluation metrics, performance of the machine learning model is not good enough to be put into real use. However, model prediction capability could be improved with some modifications concerning the data, preprocessing phase and changing the predicted categories.