Anomaly Detection of Vehicle Data Based on LOF Algorithm

Abstract

The official vehicle audit is an important issue in the government management, and it is very difficult to detect the potential doubts data in the collected data of official vehicles since most of them are unlabeled data. In this paper, a combination of DBSACN and Local Outlier Factor (LOF) algorithm is proposed for the official vehicle anomaly behavior detection by detecting the abnormal use data of the official vehicle under the same conditions. The detected data is regarded as a doubt data and submitted to the audit department for verification. Since the discrete features of the data set are too much and could not conform to the input type of the algorithm, the features are coded by One-Hot encoding, and a series of operations such as data cleaning and feature calculation are performed, and then compared with DBSCAN, LOF, and isolation forest anomaly detection algorithms. The experimental results show that the proposed algorithm outperforms the isolation forest, LOF and other machine learning algorithms in the anomaly detection of unmarked official vehicle data.