An increasing number of studies apply machine learning techniques to predict corporate fraud, and most studies claim that machine learning is more efficient than traditional regression methods in fraud detection. The primary purpose of this research is to assess the efficiency of machine learning in detecting accounting fraud. We use a sample of Chinese publicly listed firms sanctioned for accounting fraud between 2007 and 2022 and build eight models by combining powerful machine learning and SAS visualisation techniques. We find mixed results based on four commonly used performance evaluation metrics. On the one hand, unlike many recent studies, we fail to generate robust and conclusive evidence that existing machine learning techniques are more efficient than traditional logistic regression in detecting accounting fraud. On the other hand, we find that deep learning offers great potential for fraud detection. Our study contributes to a better understanding of the application of machine learning in accounting research.