A Hybrid Approach Based on Non-parametric Attribute Learning Technique and Multi-layer Networks for Congestive Heart Failure Risk Prediction

The currently emerging intelligently-driven medical diagnostic systems could offer lasting and cost-effective solutions to life-threatening diseases such as congestive heart failure (CHF). Meanwhile, developing an adequately robust CHF risk prediction method in the context of such systems is still a challenge. In this study, a novel approach driven by hierarchical neighborhood component based-learning (HNCL), a concept based on non-parametric attribute learning technique and multi-layer networks (MLN) is proposed for CHF risk prediction. In the approach, the HNCL technique intelligently extract a set of informative features and thereafter learns the interactions among the CHF attributes in other to optimize the accuracy of CHF risk prediction. In that regard, the individual contribution of the CHF attributes was obtained in the form of feature weights that was later applied to build an MLN model for the efficient prediction of CHF risk. The performance of the proposed method was tested and validated using a widely adopted clinical dataset of potential heart failure patients, and also compared to the commonly applied risk prediction method. Analysis of the obtained results shows that the proposed hybrid (HNCL-MLN) method achieved significantly higher and stable CHF risk predictions compared to the commonly applied method. The obtained results suggest that the proposed method could potentially aid the practical implementation of consistent and accurate CHF risk prediction system. It may also facilitate the development of wearable mobile health-care gadgets capable of monitoring CHF status in individuals.