Knowledge Discovery in Multivariate Clinical Time Series based on Hierarchical Structure Modeling

慢性病临床数据是一组多变量的时间序列，需要自动发现对症状识别和治疗有用的知识。本研究旨在开发慢性丙型肝炎临床检查和治疗史的建模，特征提取和症状预测方法。我们描述的症状分层应用自回归模型和自回归条件异方差模型，提取功能使用建模结果，并得出知识的基础上有多少测量是必要的症状识别的模型订单。除此之外，我们开发了一个使用核逻辑回归来预测肝损伤程度的系统，并提出了一个新的分类器，该分类器基于核逻辑回归，并使用判别学习来训练自己。

It has been required to automatically discover useful knowledge for symptom recognition and treatment in clinical data on chronic disease, which is a set of multivariate time series. This research project aims to develop methods of modeling, feature extraction, and prediction of symptoms for clinical examination and treatment histories on chronic hepatitis C. We described symptoms by hierarchically applying autoregressive models and autoregressive conditional heteroscedastic models, extracted features using the modeling results, and derived knowledge on how much measurements are necessary for symptom recognition based on the model orders. In addition to that, we developed a system to predict the degree of liver damage using kernel logistic regression and proposed a new classifier that is based on kernel logistic regression and trains itself with discriminative learning.