Research Of Association Rule Mining Algorithms And Their Applications Of Coronary Heart Disease Diagnosis And Treatment With Traditional Chinese Medicine

Coronary heart disease has become one of the major diseases that cause human death. How to mine the implicit association rules effectively from huge coronary heart disease diagnosis and treatment cases and apply the rules to the diagnosis and treatment of coronary heart disease has very important theoretical and practical significance.According to the characteristics and the mining demands of traditional Chinese medicine data of coronary heart disease, a vector-based algorithm is introduced into association rules mining to get syndrome differentiation rules from the relationship between the eighteen symptoms like pulse condition, tongue nature, predisposing factors and the twenty-six traditional Chinese medicine syndromes of coronary heart disease. And a series of coronary heart disease syndrome differentiation rules are concluded, which offer important decision-making support for diagnosing and preventing coronary heart disease.Coronary heart disease diagnosis data is often multi-labeled data with multi-value attributes, but traditional association rules mining algorithm usually mines the implicit rules directly without combining domain knowledge and has a low efficiency. Hence this dissertation disposes decision attributes and non-decision attributes by block encoding according to the feature of coronary heart disease and proposed an antecedent-consequent constraint based association rules mining algorithm. It can mine the law of medicine taking in coronary heart disease treatment in traditional Chinese medicine effectively and it is a meaningful algorithm in solving association rules mining problem which based on the relation between decision attribute and indecisive attribute.The traditional association rules mining algorithms usually mine frequent itemsets, under a unified support threshold and lead to the failure of mining long itemsets with low support. On the other hand, it will produce a lot of redundant short itemsets if a lower support threshold is provided and the efficiency of the algorithm is poor. Therefore, this dissertation proposes a new association rules mining algorithm under the condition of length-decreasing support constraint, which can mine more long patterns effectively and reduce invalid short patterns. Applying this algorithm to the diagnosis data of coronary artery disease, we can get an ideal long pattern and reduce a large number of redundant short patterns. All of these make the algorithm much more valuable in theoretical and practical point of view and it can be effective in assisting the treatment of coronary heart disease.For now, a large amount of research works have been done on the dependencies between antecedent and consequent of the association rules and some correlation measure methods have also been proposed. But most of the methods are based on the condition that the dependencies between antecedent and consequent are unchanged during the whole process of development. However the research on global correlation is of great fortuity. The support and confidence of the association rules may change along with the variation of the correlation between the itemsets as time goes by. Sometimes the rules with higher positive correlation can even become irrelevant. To overcome the shortages in global correlation, this dissertation proposes a segmented non-linear regression and backward verification method and describes its application in proving the correlation of the association rules. This method can make more accurate analysis on correlation of association rules, at the same time it can significantly reduce the amount of rules and make the rules more meaningful. The experimental results on coronary heart disease data show that this method has a very important practical significance.At present, the mining of association rules is normally based on the support-confidence framework, which is not very effective if the factors such as time and correlation of antecedent and consequent are taken into account. A novel time-validity support and time-validity match association rules framework is proposed. First, a new match measurement is defined to substitute the traditional confidence in solving the correlation problem between antecedent and consequent of rules. Moreover, by embedding the time-entropy factor into the new support-match framework, it can make the rules generated reflect the time effectiveness of the data. Then an example is given to prove the feasibility and superiority of the new method. Finally an association rules maintenance algorithm and its implication are proposed to deal with the newly added database. Experimental results and comparisons with traditional incremental method demonstrate the effectiveness of the proposed framework.