This paper designs a set of technical routes for big data auditing based on artificial intelligence, in which big data machine learning algorithms, neural networks, and NLP techniques are incorporated to analyze a large amount of audit data information; random forests and support vector machines are used to identify anomalies based on supervised and unsupervised learning, respectively, while clustering is used to further analyze the hidden risk points; and With the help of multi-layer pipeline tools to complete data cleaning, feature extraction and classification prediction and other functions. After practical verification, this system can effectively improve the quality and efficiency of audit supervision, and the recognition rate of anomalies is 22.7 percentage points higher than the traditional model. The model proposed in this paper is verified in actual cases, proving that it has advantages such as high coverage, short time and low cost, etc. It also puts forward the issues of algorithm interpretability and data security and inter-agency coordination as future research directions.
