Specific Health Examination Data Prediction for Female Subjects with Unhealthy-Level Visceral Fat Using Self-Organizing Maps

摘要

In this paper, a data-prediction method is presented for female subjects with unhealthy-level visceral fat, using self-organizing maps (SOMs). It assumes that the original data measured in successive three years are available for the subjects associated with training data for map learning. It introduces the standardization for original item values to prepare the data. When predicting original item values, the data of the subject to be explored is presented to the trained map to determine a winner. A set of the training data making that winner fire is picked up, and the differences between original data measured in the second year and that measured in the third year are calculated for each of the subjects with such training data. The proposed method adds the differences averaged over such subjects to the original data of the subject to be explored, to obtain prediction result. It is revealed that the favorable prediction accuracy is achieved for hemoglobin A1c.