From: Healthcare predictive analytics using machine learning and deep learning techniques: a survey
Methods | Advantages | Disadvantages |
---|---|---|
Convolutional neural networks (CNNs) | • CNNs excel at image classification, object detection, and segmentation. Convolution operations help them recognize image patterns • CNNs can handle image noise, making them a good choice for noisy tasks • CNNs are versatile computer vision tools for image processing | • Training CNNs requires large datasets, which may deter some users • CNNs are computationally expensive to train and deploy, which limits some applications |
Long short-term memory (LSTM) | • LSTM networks can learn long-range dependencies, making them suitable for natural language processing and speech recognition • Robust to noise: LSTM networks are good for noisy data tasks • LSTM networks are versatile ML tools | • Training LSTM networks require large datasets, which may deter some users • LSTM networks are computationally expensive to train and deploy, which may limit their use • Sometimes wrong: LSTM networks are not ideal for all tasks. For tasks with poorly defined data or nonlinear data relationships, they may not be as effective as other models |
Recurrent convolutional neural networks (RCNNs) | • RCNNs can be used for a variety of tasks, including: • Natural language processing: RCNNs can identify sentiment and entities in text • Speech recognition: RCNNs can transcribe audio and identify speakers • Time series analysis: RCNNs can predict time series values and identify patterns | • RCNNs are versatile because they can handle spatial and temporal relationships • RCNNs can learn long-range dependencies, making them suitable for natural language processing and speech recognition • Robust to noise: RCNNs are good for noisy data tasks • RCNNs are versatile ML tools |