Table 2.

ML approaches that can be used in Biomaterials and TE applications.

Algorithms	Category	Assumptions	Benefits	Limitations	Ref
Linear regression	Regression	Linearity, fixed features, independence, normality; Error variance is assumed to be constant.	Simple application; Guaranteed to find the optimal solution.	Only works for linear relationship data.	[69,70]
Random forest	Classification	Assume model errors are uncorrelated and uniform.	Provides fast learning and highly accurate predictions; Can intake large set of data without variable deletion; Can work with unbalanced data sets.	Time-consuming to form predictions.	[71,72]
Decision tree	Classification, Regression	The classes must be mutually exclusive.	Easy to use and to understand, efficient algorithm (especially for predictions).	Depending on the selection order, missing factors from the characteristic overfitting.	[71]
Neural networks	Classification, Regression	Variable independence, linearity.	Can be used for classification and regression, able to use the Boolean functions; Allows inputs with noise.	Overfitting due to too many attributes; Hard to understand the algorithm structure.	[71]
Support vector machines (SVM)	Classification, Regression	Model assumptions depend on the probability of default (PD).	Complexity of the model can be easily controlled; The models use non-linear boundaries.	Hard to understand the algorithm structure; Data training is slow.	[69,71]
Kernel ridge regression (KRR)	Regression	Linear or nonlinear function.	Computational simplicity; Prevents overfitting.	Computationally expensive.	[73,74]
Bayesian optimisation (OP)	Optimisation	A non-convex problem; No access to derivative.	Hyperparameter tuning; Cost-efficient.	The objective function can’t be modelled; High dimension problem.	[75,76]