Application of Clustering-Based Outlier Scoring with GDBSCAN for Health Care and Food Safety Domain

Abstract

Outlier detection is an essential task of discovering unusual or rare patterns in a dataset that have exceptional behavior compared with the other records in the dataset. Outliers do not follow patterns with other objects in the dataset. Many approaches identify outliers in numerical data; our focus is on identifying the noise points and cluster-based outliers in the categorical data. Due to the scarcity of labeled data, high cost, and time required in data annotation of the labeled data, it is impractical to obtain labeled datasets. The challenges led to a growing interest in the unsupervised methods, which do not require labeled data and find patterns within the data. The Density-based spatial clustering Application with Noise (DBSCAN) is integrated with the grid search for hyperparameter tuning, and a special scoring mechanism for identifying the noise points and cluster-based outliers in a categorical dataset.  The method CBOS-GDBSCAN is proposed in our earlier study.  To demonstrate the practical applicability of the method, in this study, Clustering-based Outlier Scoring with Grid DBSCAN is implemented. The method CBOS-GDBSCAN effectively identifies outliers in the dataset characterized by high-dimensional and large-scale categorical features. The method is applied to the real-world dataset taken from the UCIML repository of diverse domains such as healthcare and food safety. The experimental analysis demonstrates that the method effectively identifies outliers effectively show that the method outperforms and detects outliers in a large-scale categorical dataset.