A comparative study of anomaly detection methods for gross error detection problems

The chemical industry requires highly accurate and reliable measurements to ensure smooth operation effective monitoring of processing facilities. However, measured data inevitably contains errors from various sources. Traditionally in flow systems, reconciliation through mass balancing is applied reduce error by estimating balanced flows. this approach can only handle random errors. For non-random (called gross errors, GEs) which are caused measurement bias, instrument failures, or process leaks, among others, would return incorrect results. In recent years, many detection (GED) methods have been proposed the research community. It recognised that basic principle GED a special case outliers (or anomalies) analytics. With developments Machine Learning (ML) research, patterns be discovered provide anomalous instances. paper, we present comprehensive study application ML-based Anomaly Detection (ADMs) context on number synthetic datasets compare results with several established approaches. We also perform transformation its associated original results, as well investigate effects training size performance. One class Support Vector outperformed other ADMs five selected statistical tests for Accuracy, F1 Score, Overall Power while Interquartile Range (IQR) method obtained best selectivity outcome top 6 AMDs tests. indicate potentially problems.