Fully automatic, un-obtrusive, tool independent approach to predict the expertise level of modeler, based on the BPMN artifacts they draw.
Current implementation uses a feed-forward neural network, with one hidden layer comprising 50 neurons. The input layer contains 10 neurons, one for each feature of the vectorial representation of the model. The output layer contains 1 neuron, whose value distinguishes between the two classes (i.e., novice or expert).
The multilayer perceptron was trained with a learning rate of 0.3 and a learning momentum of 0.2. Additionally, we set the number of training epochs to 500.