Outlier Detection Based On Neural Network

In recent years, large amounts of data are generated from our daily life, and people want to find some rules through the vast amounts of data. In the process, people always think that only those who are a large number of similar data will contain some important information and ignore the small datasets. However, the datasets which people ignore may contain important information. It is the reason why people propose outlier mining algorithm.Outlier mining is also known as outlier analysis, anomaly detection, which is an important aspect of data mining. Currently, there are many outlier mining algorithms have been proposed. Most of people are familiar with distance-based and density-based algorithm, but the two algorithms have some shortcomings. They do not apply to some high-dimensional data.This paper proposes a neural network-based outlier mining algorithm. This algorithm uses neural network to classify the data and then uses the entropy to determine whether the data are abnormal. This algorithm avoids the shortcomings generated by other outlier mining algorithms. Finally, some experiments are conducted on some datasets. The experimental results show that the algorithm can effectively find the outliers.