I have an input 16821 x 13 and an output 16821 x 1. After training I have a large MSE for testing (1.1E8) but R value about 0.9. Can I consider my result to be good?
Data splitting should be done with Care. Randomly dividing the data into three disjoint sets is not enough. The foremost requirement is that the data in these subsets should have almost same Sd and mean.
Perhaps I don't understand the question.
do you have an MSE value so high in the entire dataset or only in the training set? In witch way do you split your dataset? Randomly?
I have a 5000x13 for input, 5000x1 for output. I divided the dataset into training (75%), validation (15%), Testing (15%). Number of hidden neurons 150 and the algorithm Levenberg-Marquardt. All R's are close to 1 but all MSE are high. Perhaps this algorthm cannot predict the model.
Data splitting should be done with Care. Randomly dividing the data into three disjoint sets is not enough. The foremost requirement is that the data in these subsets should have almost same Sd and mean.
Please find out mean absolute percentage error manually for sample test data and confirm with your target results. It predicted and targets are same then its fine.
- Badges
- Science topic