An Innovative Hybrid Control Chart Combining Wavelet Decomposition and Support Vector Machine for Effective Outlier Detection

Taha Hussein Ali^1*Hutheyfa Taha²Bekhal Samad Sedeeq¹Heyam A. A. Hayawi²

• ¹Salahaddin University, Erbil, Iraq
• ²University of Mosul, Mosul, Iraq

The present study proposes an innovative control chart based on integrating discrete wavelet decomposition and support vector machine (SVM) classification, which is more effective at detecting outliers during process monitoring. Common Shewhart control charts, like the X̄ chart, do not perform as effectively in the presence of outliers, generating false out-of-control signals and missing detections of problems. To overcome this, a new hybrid DWS-X̄ chart combines wavelet filtering of the subgroups' means with support vector machine (SVM) classification to distinguish between normal variation and out-of-control process events. The new chart's performance was evaluated through extensive computer simulations varying sample size, number of subgroups, and degree of outliers. Significant detection measures such as Detection Rate (DR), False Alarm Rate (FAR), and Average Run Length (ARL) were employed to assess the performance of the hybrid DWS-X̄ chart against the standard X̄ chart. The effectiveness of the charts was thereafter tested in actual clinical practice with real neonatal heart rate data. The hybrid DWS-X̄ chart outperformed the traditional X̄ chart in terms of detection rate, false alarm rate, and average run length in most cases when the data was contaminated. These findings were verified through the analysis of de-identified neonatal heart rate data, obtained with institutional approval, confirming that the hybrid chart was more efficient in detecting abnormal heart rate patterns of newborns. It indicates the potential of the proposed hybrid control chart as a valid outlier detection technique in industrial as well as healthcare process monitoring.