Outlier Detection Techniques

Introduction

Following the work of Alimohammadi [2023], given different characteristics of production time series such as trending, aperiodic, and non-autocorrelated, unsupervised outlier detection algorithms designed for time series are particularly suited for analysing production data, especially since the data unlabelled.

These algorithms can be broadly cast into statistical, regression-based, and ML groups, with multiple methods available within each category as deprecated below:

Taxonomy of unsupervised outlier detections for timeseries (more techniques exist for such purposes other than 17 shown; the above list is likely the most relevant to production data analysis)

References:

Alimohammadi, Hamzeh, A Holistic Data-Driven Framework for Forecasting & Characterization ofTight Reservoirs. University of Calgary Thesis, 2023

Introduction

See Also:

References: