Hsc-attention multi-disease prediction Net: A heuristic-assisted hybrid serial cascaded Attention-based Network with ensemble feature selection process for Multi-Disease prediction

Abstract

In the medical field, various experiments are conducted to predict disease in future healthcare using machine learning approaches. At the same time, there is a lack of a model that could not possible to predict more than one disease prediction. Here, the Electronic Health Records (EHR) helps to record the sick people’s health condition which offers special opportunities to connect healthcare events with basic disorder factors. A lot of data-driven models are emerged for health care decision-making, and thus, it provides the enhanced study on distinct learning systems to predict various disorders. The main intention of the research work is to predict the multi-disease effectively with the help of a hybrid cascaded deep learning system with heuristic development. Initially, the original data is collected from the benchmark resources. Then, it is forwarded to the ensemble feature selection model, which comprises of statistical features, optimal selected features, and deep features. Here, the Modified Stability Bound of Energy Valley Optimizer (MSB-EVO) algorithm is proposed for determining the optimal features. Once the three features are obtained, it is integrated with optimal weights that are optimized with the help of developed MSB-EVO. Then, the feature fusion is takes place. Finally, the fused features are subjected to a Hybrid Serial Cascaded Attention-based Network (HSC-AttentionNet), which is structured with a Deep Temporal Convolution Network and Long Short-Term Memory (LSTM). For further enhancement, the parameter tuning in the model is optimally selected by MSB-EVO. The performance is investigated with distinct parameters in contrast with traditional models. The several analyses are made to show the effectiveness of the developed model, which attains 96% and 95% regarding accuracy and F1-score. Throughout the analysis, the developed model shows higher performance when compared with other-state-of-art methods. Hence, the suggested model has outperformed the prediction process to easily diagnose multiple diseases and treat the patients accordingly.