Deep Learning

Data are prepared for a classification problem.

It is a supervised learning problem.

The deep learning system has to map the features vectors with the labels.

I have 340 labels and 30 features.

First of all I tried a simple logistic regression algorithm but I got better result with TensorFlow deep learning.

I used TensorBoard to compare a lot of models. In my case I have 340 labels (68 zones).

I did a lot of tries with different models. I got the best results 95% of accuracy with th following parameters:

• learning_rate=0.0001
• 12000 iterations
• 5 hidden layers of 1100

After training the model I checked the quality with th validation file. I got 95% of accuracy that confirms the model.

2 TensorBoard graphs of 2 different models