Table 5 Advantages and disadvantages of the most common algorithms used in LTBI differential diagnostic models

Support vector machine (SVM)

Good generalization ability for high dimensional and nonlinear problems;

Can adapt to different data types by selecting different kernel functions;

Performs well with a small amount of data

Long training time for large-scale datasets;

Challenging to select the appropriate kernel function and parameters for noisy data and nonlinear problems;

Does not provide direct probability estimates

Decision trees

Easy to understand and interpret;

Can handle nonlinear features and large-scale data;

Suitable for both classification and regression problems;

Minimal data preprocessing is required

Prone to overfitting, especially with deep trees;

Performs poorly with continuous and highly correlated features

Random forest

High accuracy;

Can handle high-dimensional and large-scale datasets;

Robust to noise and missing data;

Provides feature importance estimation

A more complex model with longer training time;

Substantial memory consumption for datasets with large feature spaces;

Less effective for highly correlated features

Logistic regression

Simple and fast computation;

Interpretable parameter weights to understand feature importance;

Suitable for binary classification problems

Performs poorly with nonlinear relationships in the data;

Prone to underfitting;

May not perform well with high-dimensional data or highly correlated features

Hierarchical clustering

No need to specify the number of clusters in advance;

Provides a hierarchical structure of clusters;

Works with numerical and categorical data;

Allows for visual analysis through dendrograms

High computational complexity;

Difficulty with high-dimensional data;

Restrictions on data types;

Irreversible clustering results