Table 2.

Classification and Prediction of Healthy Control and UHR’s Individuals’ Outcomes Using Spiking Neural Network.

(a) Classification accuracy outcome
Group	Control	Remitted	Maintained	Total accuracy	Standard deviation
Accuracy %	0.89	0.71	0.64	0.78	0.2

(b) Prediction accuracy outcome
Group	Control	Remitted	Maintained	Total accuracy		Standard deviation
Accuracy %	0.94	0.68	0.68	0.80		0.2

(a) Classification of 169 samples into three classes: HC (class 1, containing 81 samples), Remitter group (class 2, containing 58 samples, and maintained group (class 3, containing 30 samples. Class labels are extracted from T4 (at 24 month). For classification, the whole length of cognitive and social time series (T0-T4) was used in the training and testing sets. The results are the average of balance accuracies from 30 rounds of 2-fold cross-validation. (b) Prediction of three classes (HC, Remitted, and Maintained, labels are extracted from 24 month). For prediction, the length of cognitive time series in testing sets was 18 months (only T0-T3) to predict the outcome at month 24. This is to predict which individual is likely to remit or maintain the level of social- cognitive statues at time T4 (at 24-months of assessment) when the SNN model was only tested by the data from earlier time (18-monthss). The SNN parameters set as the following: Encoding threshold: 0.50; Firing threshold: 0.50; STDP learning rate: 0.01; Mod: 0.8; Drift: 0.005.