Evolving Deep Neural Network Architectures by Sine Cosine Algorithm for Healthcare 4.0

Prevention, early diagnostics, and timely treatment have become essential in improving patient outcomes in the area of healthcare. Furthermore, a relatively new concept, healthcare 4.0, makes use of emerging technologies to improve the overall healthcare system efficiency. However, by exploring available literature, it was noticed that the potential of deep neural networks (DNNs) has not been investigated enough in this area. For early diagnostics and timely appropriate treatment, DNNs are tasked with drawing conclusions from provided data and detecting outcomes. Despite many advantages, the DNNs need to be tuned for each particular challenge. This non-deterministic polynomial hard (NP-hard) problem is known in the literature as hyperparameters optimization. Since the metaheuristics algorithm proved to be very efficient in addressing such tasks, research conducted for the purpose of this manuscript introduces an improved variation of the sine cosine algorithm (SCA). The proposed method is used to evolve DNNs architectures and it was validated against three real-world medical datasets, including two datasets covering heart disease and one concerning liver problems. The proposed method is compared to several other contemporary metaheuristics and achieved results revealed that the introduced approach has great potential when applied in healthcare, outperforming competing algorithms in terms of classification metrics.