Analyzing the co-occurrences of allergies applying association rule mining based on nature-inspired optimization algorithms

Background and aims: While co-occurrences of autoimmune diseases have been widely studied,the co-occurrence cases of allergies have not been well-considered. Co-occurrences of allergiesmight demonstrate similarities in allergens or impacts of co-factors, furthermore, they aid in understandingthe chain of immune events leading to allergic reactions. In this way, investigating theassociation between allergies opens new ways to predict the prevalence of allergies not detectedin a human subject.Method: In this research, we utilize the data collected through a cross-sectional cohort of ۳۳۳,۲۰۰children in Philadelphia. Data includes the occurrence history of eighteen food and non-foodallergies in human subjects. We extract the co-occurrences of allergies containing ۴۸,۸۰۰ caseswith two or more allergies detected simultaneously. We define an association rule mining task utilizingan ensemble of nature-inspired optimization algorithms. In this framework, particle swarmoptimization (PSO), moth-flame optimization (MFO), grey wolf optimization (GWO), as well as,differential evolution algorithms are adopted to maximize the fitness of found rules. Top-k rulesgenerated by different methods are added to a repository of rules.Results: Our result includes a repository of association rules generated using five nature-inspiredoptimization methods corresponding to maximum fitness. Furthermore, our framework computesfive various metrics measures to evaluate the performance and confidence of detected rules.Conclusion: In this study, we proposed an ensemble framework of nature-inspired optimizationalgorithms that are utilized to mine an allergy dataset and generate association rules. This frameworkbenefitted from several highly accurate optimization algorithms in a problem with a limitednumber of features, i.e., allergies.With respect to epidemiological studies, allergies have not happened randomly. There existgroups of autoimmune and allergies that are more likely to be co-occurred. Detecting co-occurringallergies using machine learning algorithms improves the accuracy of allergy prediction andearly intervention with minimum cost. In addition, generated association rules are able to directthe co-factor and interaction research studies to start with more likely to be happened patterns.Interestingly, antecedents of some of the rules include non-occurrences of allergies which meansthere are pairs of allergies that do not occur together. We believe, however, co-occurrences ofallergies have been interesting the prevention patterns are more fascinating facts that required tobe studied.