Data Analysis Of Post-stroke Depression Scales

In this work, we applied data analysis to post-stroke depression research in order to study the prediction model of depression and key features. Main work of data analysis contained preprocessing, probabilistic risk model, classification model and association rules of the feature. The aim of probability and risk models was to forecast the depression tendency. Classification model was aim to forecast whether the stroke man was depressive. Association rules of the features were expected to mine some feature sets related to depression or stroke, which explored cause of formation of post-stroke depression.At first, this work introduced smart healthcare and data analysis. Stroke is one of the three major causes of death, and post-stroke depression is the culprits to cause treatment difficulties and worsening of stoke. The data of post-stroke depression scales was clinical data from stroke men including social demographic factors, living habits, disease factors, drug factors, genetic factors, and assessment scale, which contained 48 features. In the research of probabilistic risk model, this work mainly studied PLSR, and experimental results showed that PLSR could achieve better results in elimination multicollinearity specially. In the research of classification model, this work studied mainly FCBF, CFS, RF and GA. Moreover we proposed an algorithm named GA-RF which was a feature selecting algorithm to use GA to improve efficiency of search and wrap RF classifier. In the algorithm, classification accuracy of RF was the evaluation function values. This work designed and realized the algorithm and applied it to the research. In the research of association rules of the feature, we studied mainly Apriori and Eclat algorithm with different evaluation functions including support-confidence, lift and all-confidence, which were applied to mine frequent feature sets.