A machine learning model to predict efficacy of neoadjuvant therapy in breast cancer based on dynamic changes in systemic immunity

Comparisons among the 10 ML algorithms based on comparing accuracy and kappa: (A) boxplots presenting the accuracies and kappa values of the 5 models; (B) correlation between kappa and accuracy in the 5 models; (C) receiver operating characteristic (ROC) curves when applying the random forest model in predicting the internal validation set; (D) ROC curves when applying the random forest model in predicting the external test sets; (E) importance of features in the random forest models. Mean decreases in accuracy and gini (sorted decreasingly from top-to-bottom) of attributes were as assigned by the random forest. RF, random forest; KNNs, k-nearest neighbors; SVMs, support vector machines; CART, classification and regression tree; LDA, linear discriminant analysis; GBM, stochastic gradient boosting machine; gcvearth, multivariate adaptive regression splines; NNET, neural network; MLP, multi-layer perceptron.