07-13-2017 07:26 AM
I have a training dataset of about 30,000 observations (and a validation dataset of the exact same size) with 16 input nominal variables and 1 binary target variable. I fitted 2 models:
model1: Logistic Regression Model (using mostly the default options, see figure 1)
model2: Neural Network with one hidden node (see figure 2 for network settings, figure 3 for network structure)
I then used the "model comparison" node to compare the above models. Comparing their fit statistics, i can see that most of the values are extremely similar except one, namely "Train: Akaike's Information Criterion". Model1 and Model2 have an AIC value of 23660 and 47215 respectively. Now, I read that when comparing models that uses the same likelihood estimator, the model with the lower AIC is better. Are model1 and model2 using the same likelihood estimator? Is it correct for me to say model1 is the superior model base on the large discrepencyin AIC? Please advise, thanks in advance.