Improving Response Time Prediction for Stack Overflow Questions

This paper proposes a new machine learning model to predict the response times for questions posted on the Developer Q&A community website known as Stack Overflow. The paper first presents a comparative analysis of similar work done in the literature. Then, the paper presents a new problem formulation that represents the prediction problem accurately. Our proposed problem formulation focuses on whether or not a question will receive an answer within a certain time frame, rather than predicting a specific interval of time within which an answer should be received. The paper also introduces a new set of features that help developers better characterize the properties of posted questions. The paper apply feature engineering techniques on the raw questions posted publicly on Stack Overflow over the period of three months and addresses the problem of imbalanced dataset. The paper then uses the generated dataset to train classification models using different machine learning algorithms and evaluates the accuracy of each algorithm using cross-validation methods. Our proposed model was able to achieve a classification accuracy of 63% using feature sets extracted from the question header only.