Deep Learning in Multiple Multistep Time Series Prediction

This project aims to use deep learning to try to solve a Kaggle competition problem - Web Traffic Time Series Forecasting (https://www.kaggle.com/c/web-traffic-time-series-forecasting).

A report is generated in the PDF file. Deep Learning in Multiple Multistep Time Series Prediction

All the python code is implemented in the IPython notebook.

Note: The competiton is closed on Nov. 13th 2017. The result above achieved top 2% in the leaderboard. Great thanks to Kaggle and Google research for organizing this interesting and challenging competition.

Dataset

The dataset is from a Kaggle competition and here is the link:

https://www.kaggle.com/c/web-traffic-time-series-forecasting/data

Python libraries

• Pandas

It is mainly used to import, save, join and analyze dataframes like using read_csv, DataFrame, merge, etc.

• NumPy

It is mainly used to transform array like taking log() and exp().

• scikit-learn

Its model evaluation metrics and preprocessing functions are used.

• Keras

It is used to build neural network architecture, train and save the model, and predict new dataset like using models, layers, callbacks, etc..

• matplotlib

It is used to visualize time series and other figures.