Integration of reinforcement learning and adaptive resonance theory for data classification with application to medical diagnosis
The proposed project aims to formulate an integrated reinforcement learning (RL) and adaptive resonance theory (ART) neural network-based classification system with application to medical diagnosis. Robust RL algorithms will be developed and combined with the ART network to form an RL-ART system. The proposed project will investigate the integration of RL and ART for formulating a data classification system and to evaluate the applicability of the RL-ART system as a medical diagnosis tool. The specific objectives of the proposed project are:
- to design and develop a classification system that combines RL and ART into an integrated framework;
- to evaluate the various RL algorithms for improving the robustness of the RL-ART system;
- to study and compare the RL-ART system with other classification methods using different performance metrics (accuracy, sensitivity, specificity, and noise rejection); and
- to demonstrate the practical applicability of the RL-ART system to medical diagnosis with real (anonymous) heart attack and stroke patient records.