Skip to main content
Fig. 2 | BMC Genomics

Fig. 2

From: An improved catalogue of putative synaptic genes defined exclusively by temporal transcription profiles through an ensemble machine learning approach

Fig. 2

Comparison between the original and the improved models. Original model (above): with the whole training set we trained three algorithms: kNN, SVM and Random Forest. The hyper parameters of each classifier were set by exhaustive grid search combined with 10-fold cross validation over the training set. Finally, we increased the classification threshold of the classifiers and considered the intersection between the resulting catalogues. Improved model (below): first, we sub-sampled five times the original training set, leaving out each time a different fifth of the positive and negative examples. By this procedure we obtained five smaller, slightly different training sets. Using the positive and negative examples left out in each iteration, we created five test sets, used to independently evaluate each classifier. With each training set we trained three algorithms: kNN, SVM and Random Forest, thus obtaining 15 different classifiers. The hyper parameters of each classifier were set by exhaustive grid search combined with 10-fold cross validation over the training set. After training, we evaluated each classifier (accuracy, ROC and F1) using a different test set. Finally, we increased the classification threshold of the classifiers and considered the intersection between the resulting catalogues

Back to article page
\