Hybrid Outlier Detection in Healthcare Datasets using DNN and One Class-SVM
Roy Thomas, J. E. Judith · 2020 4th International Conference on Electronics, Communication and Aerospace Technology (ICECA) · 2020
In general, a dataset may contain a minor portion of data objects, whose properties are not identical to the majority of its data objects. Such data objects are called outliers. Most of the data mining and machine learning tasks are aimed to detect the general properties of the dataset. Outlier detection plays an important role in various application domains such as malicious activity detection in software applications, fraud detection in financial transactions, intruder detection in communication systems etc. In this paper, the performance of two major outlier detection algorithms is analyzed for healthcare applications. A hybrid model is used for detecting outliers in healthcare systems by using one-class support vector machine model and the autoencoder model is also proposed. Experimental results using the benchmark healthcare datasets show that the proposed hybrid method performs better than the existing models for detecting outliers in these datasets.