| Title |
Assessment of Information Availability in Measurement Data from Individual and Sequential-wise Perspectives, with a Focus on the Limitations of Linear Regression in Data Evaluation |
| Authors |
남윤선(Nam, Yun Seon);오현주(Oh, Hyun Ju);박형춘(Park, Hyung Choon) |
| DOI |
https://doi.org/10.12652/Ksce.2025.45.6.0801 |
| Keywords |
계측데이터; 순차적 데이터 관점; 개별 데이터 관점; 정보량; 선형회귀분석 Measurement data; Sequential-wise data; Individual-wise data; Information content; Linear regression analysis |
| Abstract |
In major civil engineering structures, various monitoring instruments are installed. The measurements obtained from these instruments provide information related to the behavior of the target system. Therefore, the engineering utility of the measurements can be evaluated in terms of the amount of behavior-related information they contain. The information content of the measurements may differ depending on the data processing method, even when the measurements themselves are identical. Measurement data can be analyzed from two perspectives: the individual-wise perspective and the sequential-wise perspective. The individual-wise perspective corresponds to conventional data processing methods such as linear regression analysis, whereas the sequential-wise perspective corresponds to processing approaches employed in models such as LSTM or HMM. In this study, we propose a method for evaluating the usable information content of measurements from both perspectives. Through numerical simulations and applications to actual monitoring data, it was shown that the sequential-wise perspective enables more effective utilization of the information content, whereas the individual-wise perspective can only utilize a portion of the available information, depending on the case. Furthermore, by applying both numerical simulations and actual monitoring data, we evaluated the performance of a representative individual-wise method, the linear regression model, in terms of usable information content and predictive performance. The results demonstrated that the linear regression model exhibits unavoidable performance limitations under certain conditions, highlighting its inherent constraints compared with sequential-wise approaches. |