Research And Prediction Application Of Temporal Pattern Mining Technology Based On Medical Data

It is an importance issue in biomedical informatics to explore the temporal relationship between events in Electronic Medical Records(EMR),the result of which can reveal the patient’s disease situation.Therefore,the data mining of EMR data has a very broad research prospect,which can provide diagnosis and treatment decision support for medical staff.Different from mining sequence patterns from interval-based event in the past,this paper optimizes the sequential pattern mining algorithm by using point time nominal medical event,and our algorithm is significantly improved compared with the past.We treated those mined patterns as additional features and evaluate them in a predictive modeling task based on a real world EMR data warehouse and obtained a better performance.The main work includes:(1)An algorithm for mining temporal patterns is proposed.We optimized the sequential pattern mining algorithm by using point time nominal medical event,and we proposed a time-constrained sequential pattern mining algorithm based on the bitmap,which obtained a higher mining efficiency.(2)Constructed a framework of mining sequential frequent patterns based on medical data.Based on the point time nominal medical events,we proposed a temporal pattern mining pipeline,including sequence constructor,sequence preprocessor,sequential pattern miner and Bag-of-Pattern vector constructor,which had a better mining efficiency for complicated data sets.(3)A prediction model was made to predict the onset risk of disease on a realworld data warehouse.We treated the frequent patterns as additional features to predict and evaluate the risk of congestive heart failure.We found that the prediction performance can be significantly improved by adding these sequence pattern features as additional input of machine learning prediction model.