Construction of Non-invasive Pre-diagnostic Index System for Diabetes Based on CACP Multilayer Screening Framework

Abstract Diabetes is becoming more and more prevalent, and preventing its occurrence has become a major challenge for society. In response, this article has established a non-invasive pre-diagnostic indicator system for diabetes using the CACP multi-layer data processing framework. The system consists of four main layers: cleaning, analysis, cross-validation, and processing. In the cleaning layer, indicators and subject groups with a high missing rate undergo screening, and missing data is processed through interpolation. The analysis layer analyzes the correlations between indicators of different types and categories, verifies and eliminates highly correlated indicators, and obtains a preliminary screening indicator set. In the cross-validation layer, the primary screening indicator set undergoes further processing to obtain a secondary indicator set. In the processing layer, the normalized and processed data is input into the xgboost classifier to obtain the AUC value and accuracy as a reference. Each indicator in each secondary indicator set is then deleted and processed to select the group of indicators with better results than the control group. Finally, the selected indicators are validated through the processing layer to obtain the best classification performance in the final screening feature set. Using part of the features and classification data from the Nhanes data in 2017-2018, the non-invasive pre-diagnostic indicator system for diabetes can achieve a classification accuracy of 0.91. This system only requires routine physical examination data such as height, weight, blood pressure, and some personal information to achieve autonomous non-invasive pre-diagnosis, making it an effective tool to prevent and treat diabetes.