Table 4. Ten-fold cross validation of Elman neural network (2 hidden layers with 10 and 5 neurons in each layer) run on a dataset with selected features (with 27 features) presenting overall, mesostable, and thermostable prediction accuracies.

Run	Size of training set	Size of test set	Per cent of accuracy in predicting thermostable proteins	Per cent of accuracy in predicting mesostable proteins	Per cent of Overall accuracy
1	1851	206	0.81	0.85	0.83
2	1851	206	0.76	0.92	0.86
3	1851	206	0.89	0.91	0.90
4	1851	206	0.89	0.96	0.93
5	1851	206	0.90	1.00	0.96
6	1851	206	0.74	0.99	0.89
7	1851	206	0.89	0.96	0.93
8	1851	206	0.84	0.97	0.92
9	1851	206	0.84	0.98	0.92
10	1854	203	0.88	0.93	0.91
Average			0.84	0.95	0.91