In general, ElemNet exhibits higher impact of training dataset size compared to the Random Forest models.
Metrics on computational time versus dataset size are also presented.
Our results demonstrate that deep learning models can not only benefit more with an increase in dataset size compared to traditional ML models, but also deep learning can outperform them even at relatively smaller dataset size of around 4k samples.
We augmented the image data by spatial transformations to increase the dataset size as illustrated in Fig. 1b.
We have found a positive correlation between dataset size and performance.
The error curve has a steeper reduction in test error with the increase in training dataset size in the DNN model compared to Random Forest models.
However, the important observation is that deep learning performs better than the Random Forest models even when the training dataset size is in ~103 104.
Although, as expected, the performance of every method declined with decreasing dataset size, the magnitude and timing of this decline varied strongly per method.
Previous studies detected TP53 mutations in 3 8% of myeloma patients, varying by dataset size and technique used [15, 16, 32].
Because of the relatively small dataset size (200 proteins), the method was cross-validated, instead of creating separate training and test sets.
The test dataset size was smaller than the training dataset, which involved 108,000 (1500 × 36 × 2) input vectors with labels for the 6 × 6 lattice.
