Detecting asthma control level using feature-based time series classification

Abstract

Asthma symptom control is the best method for asthma treatment. Early detection of asthma control status can provide the time required for presenting preventive treatment programs to reduce the future risk of asthma exacerbation and episodes of poor asthma control. Accordingly, the current study aims to outline a novel model for classifying asthma control status using a time series/time sequence-based classification approach. Few researchers have addressed the problem of classifying asthma control level in the context of the time series/time sequences dynamics-based approach. It is a significant defect in chronic disease management such as asthma that requires continuous monitoring of symptoms. As a result, it is needed to examine the effect of the time-series/time-sequences of the stimuli affecting asthma status in classifying the control level of this disease. We have designed a daily asthma self-monitoring form and collected a total of 2870 daily assessments of asthma control on 96 asthmatics patients older than five years for 9 months. Using a 7-day window of time, we created the asthma control dataset with clinical variables, the patient’s medical history, and environmental parameters, to detect asthma control level using a time series/time sequence-based classification model. Our best model yielded a recall of 87%, a specificity of 94%, a precision of 89%, a negative predictive value of 93%, an accuracy of 92%, an F-measure of 88%, and an AUC of 87%. Our study indicated that time series classification models among data mining techniques have great potential in creating a prognostic model for chronic diseases such as asthma. Furthermore, the time-series/time sequences of daily asthma symptoms also play a significant role in classifying asthma control levels.