Study On The Clinical Classification Analysis Methods Based On Complex Network Features

The diagnosis framework based on human phenotypes is a complex decision problem, which incorporate complicated types of information entities and their interactions. For example, Physiological time series data is an important data resource of human health, but the traditional time series analyzing way aims to discover partial structure without considering the whole information rule and mode of human’s complex system. So it has a certain Limitation. Clinical syndrome prediction which is one of the main study contents of TCM syndrome is a multi-label classification problem. A multi-label classification problem regards syndromes as independent individuals by Label split way, while traditional Chinese medicine theory believes there is the complex relationship among syndromes. Moreover, the body of the syndrome differentiation is the clinician, and the process of it can be influenced by the clinician’s experience and other factors. So it has something subjective in getting symptom signs, which lead to the serious omission of symptoms data.This paper will research the above two questions by machine learning method based on characteristics of complex network.(1)The adjustment of human physiological system to physiological signals makes its time series data fractal features. Moreover, different adjustment ability for different healthy condition makes the fractal features of it different. In the paper, we uses complex network to research physiologic time series, uses The Visibility Graph to make the data of physiologic time series network, regards the network characteristics of time series as the attribute, and regards the healthy condition from physiologic data as the class in order to analyze network features related with healthy condition. The paper analyzed the old’s physiological data collected by Standard heart rate time series data and wearable monitoring equipment CIM and found out that graph density is the main network feature related with healthy condition.(2)Based on Co-occurrence, the subjective trouble the clinicians have and clinical data samples, we respectively established symptoms network and syndromes network. In symptoms network, we employed DIAMOND Algorithm based on the network structure to enlarge sample’symptoms and employed Prince Algorithm based on transmission network structure and the model based on the hypergraph to predict symptoms. The result is that the combination of DIAMOND Algorithm and Prince Algorithm can reach a better predictive effect with the Hamming Loss 0.0925 and Ranking Loss 0.2018, while the predictive effect of the modal based on the hypergraph is worse. The result show that the relationship of syndromes and the relationship of symptoms are the inside factors of Syndrome prediction.