Algorithm Optimization and Fraud Feature Mining in Audit Data Analysis

In the analysis of audit data, algorithm optimization and fraud feature mining are important means to improve audit efficiency and accuracy. In this paper, an improved K-means clustering algorithm based on density (DBK-means) is proposed. By calculating the density of data points, the initial centroid is selected, the weighted distance measurement method is introduced, and the position and quantity of centroid are dynamically adjusted to adapt to the actual distribution of data. The experimental results show that the DBK-means algorithm is superior to the traditional K means algorithm in accuracy and recall, and it can find the hidden structure and grouping in data more accurately. In addition, this paper also combines association rule mining and anomaly detection technology to mine a series of financial indicators or transaction patterns that are highly related to fraud, such as accounts receivable turnover rate and inventory turnover rate. These fraud characteristics help auditors to identify and prevent fraud more effectively.