MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

A Comparison of Prediction Models in Autocorrelated Processes for Quality Assurance

j Lah1*, G Kim2, D Shin3, (1) Myongji Hospital, Hanyang University College of Medicine, Goyang-si, KR, (2) University of California, San Diego, La Jolla, CA,(3) National Cancer Center, Goyang-si, KR

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: We expect that the forward direction of quality improvement will enable automatic process control, such as the machine performance check (MPC). The approach used in automatic process control is to predict the next observation, and then use the mechanism to adjust so that the observation will be closer to the desired target of QA. With the growing demand for automatic QA in radiation therapy, process characteristics may present various types of dependencies in time series and data are more likely to be autocorrelated. We compared the accuracy of predictive models for autocorrelated QA data using the machine learning method, artificial neural networks (ANNs) and the traditional approach, the autoregressive integrated moving average (ARIMA).

Methods: In this study, sets of data with different patterns (non-autocorrelation and autocorrelation) were deployed to compare the performance of popular predictive models, ANNs and ARIMA. This aspect was crucial because it might enhance the predicting capability by utilizing autocorrelation as a basis.

Results: The results indicated that the ANNs is a more powerful and accurate predictive quality than ARIMA in daily output. The ANNs is effective for detecting autocorrelation and provides a prediction of the QA process average will be taken at the next time. This signified that the autocorrelation structure of QA data has no effect on the performance of the ANNs model. Although the ARIMA model was based on the autocorrelation structure, it still had higher mean absolute error (MAE) than the ANNs.

Conclusion: The autocorrelation can significantly affect the accuracy and overall performance of the predictive QA system. The nature of QA processes might cause difficulties in predicting for the future target because of its complicated structure. The predictive maintenance provides a new perspective on QA strategies to achieve a maximum life expectancy of machines while minimizing the risk of failure.

Funding Support, Disclosures, and Conflict of Interest: This research was supported by Basic Science Research Program through the National Research Foundation of Korea(NRF) funded by the Ministry of Education(Grant No. NRF-2018R1D1A1A02085342).

Keywords

Preventative Maintenance, Autocorrelation Function, Statistical Analysis

Taxonomy

IM- Dataset Analysis/Biomathematics: Machine learning

Contact Email