Mixed data outlier detection based on information entropy integration strategy

Outlier detection for mixed data is one of the key research areas in data mining. Existing ensemble strategies typically combine outlier information from various data types to assess the overall outlier degree of a data object. However, these strategies often fail to adequately consider the differences in outlier information inherent in different data types, which affects detection performance. To address this issue, a new outlier detection method based on an entropy-based ensemble strategy is proposed. First, the entropy of each attribute is used to quantify the outlier information it contains, and the outlier information from different attribute types is combined. Next, the outlier information weights for each data type are computed based on its respective outlier information, and the outlier degree of the data object is aggregated with these weights. This approach effectively reflects the differences in outlier information across data types. Finally, the proposed algorithm is validated using the UCI mixed datasets, with experimental results demonstrating that it captures the outlier information differences across data types and improves detection performance.