Hello all,
I have a dataset of 105 records, with 4 attributes. All the attributes take whole numbers as its values and three of them are predictors. I train 80% of the data with a Multilayer perceptron (MLP) as well as with a decision tree. When I measure the performance of these two models by testing the remaining 20% of data, I found that the decision tree has Mean Squared error (MSE) around 0.45 and the MLP has around 1.05. Why is that so?
I have tried changing the number of hidden layers and number of perceptrons in each layer and number of iterations, but 1.05 is the lowest MSE that I could achieve. Implementation is carried out in python with the sklearn package.
Can anyone help me?
Thanks in advance
Of course, why not? Multilayer perceptron doesn't have to perform perfectly all the time. That said, Multilayer perceptron works decently in certain circumstance, so you need to read about it and know your data as well.
HTH.
Samer
MLPs are sub-optimal sometimes and the optimality always depends on the training and testing of the data. Better consider the hybrid model of decision tree and MLP.
Thanks for your inputs Sameer Sarsam, Mahboobeh Parsapoor, Milad Janalipour and Naveen Kumar Boiroju.
- Badges
- Science topic