This paper presents the results of a systematic literature review of the field of machine learning (ML)-based bug report analysis. We reviewed 1,825 papers and analyzed 204 of them in depth. We summarize the results of the analysis into seven major findings: the utilization of algorithms such as CNN, LSTM, and kNN; feature representation using Word2Vec and TF-IDF; preprocessing methods; software projects to be evaluated; key analysis tasks; and evaluation metrics. We also suggest six directions for future research. In particular, we point out the lack of advanced models such as BERT and the absence of rigorous statistical testing.
