Table 3.

Strengths and weaknesses of proposed machine learning algorithms

Algorithm	Strength	Weakness
Artificial neural network	Could access several training algorithms [72]	The nature of being a black box [72], overtraining [73]
Support vector machine	Can avoid overfitting and defining a convex optimization problem [72]	Choice of the kernel as well as speed and size of training and testing sets [72]
Compartmental models (SIR, SEIR, SIRD, etc.)	Predict how the disease spreads Present the effects of public health interventions on the outcome of the pandemic [74–78]	The proposed models are mostly deterministic and work with large populations [79]
Nature-inspired algorithms (genetic programming)	Intelligent search [80] Can integrate with certain decomposition algorithms [81]	Several parameters should be set by the decision-makers The algorithms are approximate and usually nondeterministic [82]
Prophet algorithm	Are robust in dealing with missing data [70]	It is hard to use the algorithm for Multiplicative models Predefined format is needed for data before using the algorithm
ARIMA	Works for seasonal and nonseasonal models Outliers can be handled well	Changes in observations and changes in model specification make the model unstable [83]
Deep learning	Results comparable to human expert performance [84, 85]	Requires large amounts of data The training process is expensive