Future of Machine Learning (ML) and Deep Learning (DL) in healthcare monitoring system

Abstract

Prediction and early detection of diseases have been an important field of research for a long time to diagnose any disease. Machine-learning (ML) algorithms have proved quite efficient in disease detection and decision making in healthcare. While most ML algorithms achieved good accuracy, domain adaptation and robustness are still a big concern today. Some algorithms acquire good accuracy on some datasets while failing to do well on all datasets. These algorithms' performance can often change with some data variation in the future, so updated ML algorithms are necessary. We discuss state-of-art ML algorithms based on their accuracy, robustness, and domain adaptation, which are helpful in several healthcare datasets. In the age of Industrial 4.0, telemedicine is an important area of research. Several aged people cannot appear in hospitals to diagnose and detect diseases. Their diagnosis and care are needed to be carried out at home. So portable devices for detection and good-medicine analysis are crucial in future decades. ML are deployed for accurate detection of disease and treatment consultation based on their medical condition. Several forms of data are used for analysis, such as heart rhythm, oxygen content in the blood, and DNA sequences. Sometimes, visual recognition has played an essential role in detecting disease like skin cancer and smallpox. The convolution operation in several deep-learning algorithms has been proved as a vital feature extractor for images. We prefer recurrent neural networks because they memorize their past line and predict future values when we get data sequences. ML algorithms are pretty data-dependent, so a comparative analysis of this ML algorithm based on the data type is necessary before real life. The second reason why comparative analysis of ML algorithms is essential is domain adaptation. ML algorithms can be fooled through adversarial attacks. So, we try to approach a better way to solve a problem and make an ML algorithm more robust to hostile attacks. Some ML algorithms work best when the feature extraction algorithm provides better virtual data for classification. Expanding the dimensions to a higher value has helped differentiate the class better. In contrast, when you raise the extent, visual analysis of the data and the classification becomes a significant tissue. On the other hand, there is a question about what is happening inside the ML algorithm or in the neural network, making us more uncertain about the ML algorithms' prediction. So, we give a comparative analysis of each ML algorithm with so many issues. Thus, in this research, we proposed our model with a better approach that gains higher accuracy and confidence than traditional ML, where one solely relies on the loss of accuracy of the model. We have used breast cancer and the Pima Indian diabetes dataset (PIDD) to test our approach, and with ensemble learning, we develop confidence in our models' reliability. © 2023 The Institute of Electrical and Electronics Engineers, Inc.