Combining Naive Bayes Classifiers with Temporal Association Rules for Coronary Heart Disease Diagnosis

Abstract

We present a Naive Bayes classification model where the features are temporal association rules (TARs) annotated with their possible recurrence patterns, referred to as periodic TARs. To analyze clinical time series we rely on several Temporal Data Mining (TDM) methods, like temporal abstractions (TAs). We used this approach to diagnose coronary heart disease (CHD) based on patient history. Firstly, we exploited TAs to preprocess data and obtain qualitative and trend temporal patterns. Secondly, we applied a temporal pattern mining algorithm able to detect TARs by identifying the most frequent temporal relationships among the TAs. Finally, the classifier incorporates periodic TARs as its features, by considering the possible recurrence patterns of each TAR within the relevant patient history. A key claim of this research is that where long time periods are of significance in some medical domain, such as the CHD domain, higher order temporal abstractions can yield better performance. The viability of this claim is demonstrated by comparing the performance of the classifier with periodic TARs as its features with that of a baseline classifier whose features are simple TARs representing the occurrence or not of temporal relationships, without consideration of possible periodic occurrences of the given temporal relationships. The results obtained illustrate the comparatively high performance of the periodic TAR classifier over the baseline, simple TARs classifier, thus demonstrating the effectiveness of the proposed approach.