Detecting Online Outlier for Data Streams using Recursive Residual

Outlier analysis is a statistical procedure that involves the identification of anomalous observations. Outlier detection is important in many fields and involved in numerous applications since if the outliers are not detected then it can lead us to a wrong decision. There are a wide variety of techniques that can be used to identify outliers in datasets. Most of the existing methods are based on static data, while in case of data stream these methods are computational inefficiency and require big memory storage. This is due to the data stream being a dataset that is generated in a real-time data stream and requires real-time data analysis. The recursive least squares approach is mainly used in many regression analysis and the outlier detection process is referred to as supervised outlier detection. In this paper, we propose an online outlier detection algorithm using recursive residuals via recursive least squares method. The recursive residuals are sequentially calculated on the distance of the observation and the prediction values. To estimate online parameters in this research, we apply a recursive least squares method to predict an iterative model. In recursive least squares method a single new data point is analyzed each algorithm iteration to improve the estimation of our model parameters. Then, our simulation is performed on historical data for Tesla stock for a given period. Results that our online algorithm can demonstrate its ability to maintain high accuracy and find better outliers with low computational cost for anomaly detection.