Table 1.

Principle, advantages and drawbacks of common machine learning algorithms.

Machine Learning Algorithm	Principle	Advantages	Drawbacks
Linear regression	It assumes a linear relationship between input variables and output and thus, attempts to model this relationship by fitting a linear equation to the observed data There are several implementations of this model, of which the most commonly used is ordinary least squares, which tends to minimize the residual sum of the squares between the observed and predicted targets.	Simplicity Ease of implementation	It assumes that the input variables are independent It risks generating biased models due to oversimplification
Linear discriminant analysis (LDA)	It is used to identify to which class samples belong to, certain statistical properties of the data are first calculated and then substituted into the LDA equation. The statistical properties consist of the mean and variance for the case of a single input and the means and covariance matrix for multiple inputs.	Simplicity Robust and interpretable classification results	Does not perform well when the discriminant information is not present in the mean It cannot be applied to non-linear problems
Random Forest	It builds a number of decision trees on bootstrapped training sets and considers a random sample of m predictors to be split candidates from the full set of p predictors to overcome the problem of high variance. Therefore, on average, the strong predictor is not considered and other predictors have a better chance. This process can be thought of as decorrelating of the trees, thereby making the average of the resulting trees less variable and hence more accurate and reliable.	Reduced variation. Accurate and reliable It works well for both classification and regression problems	It requires considerable computational power and time for training It suffers from interpretability
Support vector machine	It converts a non-linear separable problem by transforming it onto another higher dimensional space and thus, the problem becomes linearly separable. This is accomplished using various types of so-called kernel functions. Then, classification is performed by finding the hyperplane that well separates the classes of samples.	It can solve any complex problem with the appropriate kernel function Less risk of overfitting	Choosing the appropriate kernel function is not easy It does not work well with large or noisy datasets
Discriminant analysis via mixed integer programming (DAMIP)	It is a classification model based on a very powerful supervised-learning approach used primarily in the biomedical field. It is a discrete support vector machine coupled with a powerful embedded feature-selection module [176].	It reduces noise and errors. It applies constraints that result in superior classification accuracy Universally consistent. Handles well imbalanced data	This algorithm is mainly used in the biomedical field, little is known about its drawbacks in literature