DAAR: Drift Adaption and Alternatives Ranking approach for interpretable clinical decision support systems

Diagnosing allergy comorbidities is challenging because the comprehensive studies on the prevalence of each allergy and the relation with other diseases are scarce. Moreover, there is a chance for uncertainty in the intradermal skin test results due to poor administration, manual observation and interpretation. Hence this work aims to develop a Decision Support System (DSS) for enhancing the diagnosis of allergy comorbidities. The DSS framework consists of a knowledge engineering phase and decision-making phase. The former focuses on detecting the concept drift, pre-processing the data, and extracting the knowledge, whereas, the latter focuses on the DAAR approach for adapting the concept drift and ranking the alternatives. The intradermal skin test data collected from allergy testing centre in India during 2012 and 2019 is used for developing and evaluating the DSS. The three techniques of concept drift adaption, namely, replacement, ensemble, and incremental are experimented on both balanced and unbalanced datasets. It can be observed from results that the incremental adaption technique has achieved an accuracy of 70.28% and 54.94% with complete decision, and 91.42% and 76.37% with partial decision on query data of 2012 and 2019 respectively. Moreover, the accuracy of the incremental technique is above the average of the accuracies of state-of-art machine learning algorithms. This implies that the framework is capable of supporting clinicians in choosing an optimal decision from the preference order of alternatives. Moreover, the interpretable knowledge provided by the system about all possible combinations of alternatives aids clinicians in understanding the reasoning behind their choice.

Automated summary

This work aims to develop a Decision Support System (DSS) for enhancing the diagnosis of allergy comorbidities. The DSS framework consists of a knowledge engineering phase and decision-making phase. The system is capable of supporting clinicians in choosing an optimal decision and providing interpretable knowledge about alternatives.