Table 1.

Performance of several models trained using classical machine learning methods and federated learning methods, where the number of participating clients in the federation is n = 2, tested on the PPMI dataset

Algorithm name	ROC-AUC	AUC-PR	Balanced accuracy	Precision	Recall	F 0.5	F 1	F 2	Log loss	Matthews correlation coefficient
AdaBoost classifier	0.865 ± 0.033	0.939 ± 0.01	0.736 ± 0.069	0.844 ± 0.043	0.89 ± 0.047	0.897 ± 0.017	0.885 ± 0.027	0.942 ± 0.013	0.632 ± 0.007	0.499 ± 0.122
Bagging classifier	0.82 ± 0.05	0.916 ± 0.025	0.69 ± 0.052	0.813 ± 0.027	0.897 ± 0.061	0.865 ± 0.021	0.862 ± 0.024	0.933 ± 0.01	1.001 ± 0.382	0.428 ± 0.123
GradientBoosting classifier	0.879 ± 0.046	0.943 ± 0.026	0.723 ± 0.072	0.833 ± 0.042	0.911 ± 0.021	0.916 ± 0.03∗	0.894 ± 0.022	0.942 ± 0.015∗	0.444 ± 0.099	0.486 ± 0.119
KNeighbors classifier	0.61 ± 0.099	0.806 ± 0.065	0.533 ± 0.029	0.729 ± 0.014	0.937 ± 0.046	0.782 ± 0.023	0.837 ± 0.007	0.927 ± 0.004	2.836 ± 0.617	0.111 ± 0.104
LinearDiscriminantAnalysis classifier	0.763 ± 0.045	0.883 ± 0.031	0.681 ± 0.053	0.826 ± 0.04	0.77 ± 0.05	0.7 ± 0.344	0.714 ± 0.35	0.776 ± 0.38	1.608 ± 0.488	0.347 ± 0.095
LogisticRegression classifier	0.831 ± 0.068	0.915 ± 0.039	0.734 ± 0.072	0.841 ± 0.043	0.894 ± 0.028	0.872 ± 0.047	0.883 ± 0.033	0.939 ± 0.011	0.648 ± 0.203	0.493 ± 0.134
MLP classifier	0.739 ± 0.078	0.892 ± 0.032	0.703 ± 0.059	0.833 ± 0.038	0.815 ± 0.054	0.843 ± 0.038	0.858 ± 0.034	0.932 ± 0.013	6.616 ± 1.844	0.402 ± 0.119
QuadraticDiscriminantAnalysis classifier	0.504 ± 0.057	0.774 ± 0.029	0.504 ± 0.057	0.725 ± 0.055	0.385 ± 0.081	0.757 ± 0.008	0.833 ± 0.006	0.926 ± 0.003	19.674 ± 1.492	0.009 ± 0.105
RandomForest	0.816 ± 0.076	0.917 ± 0.027	0.552 ± 0.034	0.736 ± 0.016	0.993 ± 0.017∗	0.857 ± 0.043	0.874 ± 0.032	0.942 ± 0.014	0.508 ± 0.029	0.249 ± 0.121
SGD classifier	0.755 ± 0.065	0.907 ± 0.025	0.735 ± 0.062	0.846 ± 0.032	0.857 ± 0.068	0.857 ± 0.037	0.876 ± 0.036	0.936 ± 0.015	7.525 ± 2.282	0.481 ± 0.143
SVC classifier	0.838 ± 0.069	0.924 ± 0.032	0.711 ± 0.071	0.827 ± 0.041	0.883 ± 0.042	0.872 ± 0.042	0.886 ± 0.024	0.941 ± 0.008	0.44 ± 0.082∗	0.447 ± 0.145
XGBoost classifier	0.89 ± 0.046∗	0.953 ± 0.018∗	0.765 ± 0.097	0.86 ± 0.062	0.911 ± 0.03	0.915 ± 0.03	0.900 ± 0.033∗	0.942 ± 0.014	0.461 ± 0.135	0.557 ± 0.167
XGBoost random forest classifier	0.857 ± 0.064	0.936 ± 0.029	0.773 ± 0.057∗	0.868 ± 0.04∗	0.885 ± 0.047	0.907 ± 0.039	0.891 ± 0.041	0.936 ± 0.011	1.79 ± 0.853	0.558 ± 0.105∗
FedAvg LR	0.69 ± 0.16	0.874 ± 0.042	0.617 ± 0.109	0.772 ± 0.054	0.955 ± 0.037∗	0.818 ± 0.054	0.863 ± 0.026	0.935 ± 0.008	0.655 ± 0.14	0.278 ± 0.25
FedAvg MLP	0.76 ± 0.102	0.872 ± 0.072	0.671 ± 0.087	0.817 ± 0.051	0.768 ± 0.089	0.708 ± 0.35	0.728 ± 0.358	0.779 ± 0.382	0.767 ± 0.308	0.334 ± 0.179
FedAvg SGD	0.828 ± 0.048	0.92 ± 0.025	0.757 ± 0.048∗	0.904 ± 0.049∗	0.707 ± 0.033	0.871 ± 0.032	0.872 ± 0.018	0.939 ± 0.008	0.545 ± 0.032∗	0.47 ± 0.084
FedAvg XGBRF	0.829 ± 0.023∗	0.924 ± 0.015∗	0.739 ± 0.058	0.848 ± 0.043	0.883 ± 0.036	0.886 ± 0.02∗	0.875 ± 0.012	0.929 ± 0.005	0.691 ± 0.0	0.497 ± 0.089∗
FedProx μ = 0.5 LR	0.755 ± 0.142	0.887 ± 0.041	0.653 ± 0.088	0.791 ± 0.042	0.941 ± 0.031	0.704 ± 0.349	0.729 ± 0.358	0.784 ± 0.384	0.609 ± 0.155	0.362 ± 0.198
FedProx μ = 0.5 MLP	0.757 ± 0.096	0.872 ± 0.061	0.695 ± 0.088	0.829 ± 0.048	0.808 ± 0.075	0.843 ± 0.042	0.868 ± 0.028	0.937 ± 0.004	0.976 ± 0.314	0.387 ± 0.182
FedProx μ = 2 LR	0.812 ± 0.079	0.906 ± 0.04	0.658 ± 0.028	0.79 ± 0.014	0.937 ± 0.025	0.866 ± 0.045	0.879 ± 0.025∗	0.941 ± 0.006∗	0.582 ± 0.137	0.398 ± 0.069
FedProx μ = 2 MLP	0.765 ± 0.079	0.868 ± 0.06	0.694 ± 0.069	0.83 ± 0.042	0.798 ± 0.045	0.706 ± 0.348	0.724 ± 0.355	0.781 ± 0.382	0.9 ± 0.368	0.379 ± 0.133

Data reported are mean and standard deviation across K = 6-fold cross-validation. Best performing algorithms for each metric are indicated by an asterisk.