A Novel Harris Hawk Algorithm (HHA) to Optimize Deep Autoencoder (DAE) Method of Deep Learning for Cancer Diagnosis

Over the past decade, the significance of data analytics in health informatics has surged due to the overwhelming influx of multi-modality data. Artificial neural networks have emerged as one of the most important and powerful influences in machine learning, rooted in deep learning. The exact prognostic of cancer remains a challenging problem for healthcare practitioners, even though cancer research has been a steady emphasis throughout the last decades. To increase a patient's chances of survival, early diagnosis is crucial, effectively differentiating between benign and malignant tumors is a considerable challenge. Deep learning has become widely accepted tool among medical researchers as an effective method in this area. A deep autoencoder architecture which was primarily designed for feature extraction is improved by using the Harris Hawk algorithm. Surprisingly, this approach comprises a pre-training step rather than requiring labelled data for training. During this pre-training stage, the proposed Harris Hawk algorithm makes use of exploration, exploitation, and attacking techniques, increasing feature extraction accuracy significantly to an outstanding 93%. This in turn greatly improves the performance of deep auto encoder applications, including those for cell grouping, 3D- brain reconstruction, cancer detection, and the prediction of demographic data. The deep autoencoder optimized using the Harris Hawk method stood out as the obvious choice in terms of accuracy when compared to other autoencoder models including sparse autoencoders (SAE), denoising autoencoders (DAE), contractive autoencoders (CAE)and convolutional autoencoders (CNAE).