AI - one size fits all?

Abstract

The use of artificial intelligence (AI) in medicine requires a careful selection of suitable models, as there is no universal "one size fits all" method. While linear regression is convincing due to its simplicity and interpretability, it is limited due to the assumption of linearity and susceptibility to multicollinearity and outliers. More complex approaches such as neural networks show their strengths in the detection of non-linear patterns and automatic feature extraction but require large amounts of data, high computing capacity, and suffer from limited explainability. Principal component analysis (PCA) offers an efficient reduction of dimensionality. Ultimately, the choice of model depends on the balance between accuracy, interpretability, and data availability. A selection of machine learning models is presented in this article.

Keywords: AI; PCA; artificial intelligence; machine learning; regression.

Figure 1. Schematic classification of various machine learning models. Division into supervised and unsupervised learning.

Figure 2. Example of a decision tree. Classification of atopic dermatitis by age and severity.

Figure 3. Example of a principal component analysis (PCA) based on transcriptome data from skin samples of patients with atopic dermatitis and healthy controls. L = lesional; NL = non-lesional; NN = healthy skin.