Wikipedia traffic prediction

This is a competition from the website kaggle.com (https://www.kaggle.com/c/web-traffic-time-series-forecasting)

The competition wants you to forecast future web traffic for approximately 145,000 Wikipedia articles.
The traffic of all the pages in the previous few years on daily basis is given. The forecast of the next 60 days is asked.

Here below the outfit of the data from a single wikipedia article:

Such a large number of pages introduces a large constrain on the complexity of the prediction.

Three models are taken into account and tested separately:

• Prophet

• DNN with a VGG16 1D configuration

• DNN standard with 5 hidden layer

The last 2 monthes have been used to forecast the next 2.

All the models have been used to predict the entire future time frame without breaking it into smaller windorws.

Submission

My result has been submited at the competition:

Ranked 100 over 400 participant (SMAPE 41.96493)

The code can be found in my github repository:

https://github.com/canada87/web_traffic