The competition allows you to compete on a selection of 2 datasets of different size but similar complexity. Dataset A is a complete dataset of 111 monthly time series drawn from homogeneous population of empirical business time series. Dataset B is a sub sample of 11 time series from the 111 time series, and is therefore contained in the larger dataset. We particularly encourage submissions for the dataset A with 111 time series. Both datasets can be downloaded following the NN5 Instructions
Dataset & Time Series Structure
The time series provided originate from different cash machines at different, randomly selected locations within England and are not related. All time series start on March 18, 1996 and run until March 22, 1998, providing two years of daily data. The objective is to predict the next 56 days (i.e. 8 weeks) into the future. The time series show different patterns of single or multiple overlying seasonality, including day of the week effects, week or month in the year effects, calendar effects that are fixed to a specific date (e.g. Christmas), calendar effects that are linked to a public holiday that may move around the year (e.g. Easter) etc. They contain both observations of "0" withdrawals, which indicate that no money was withdrawn at that day, and missing observations which are blank " " without any values, which indicate days in which no value was recorded for various reasons. No missing values exist in the test set, but 0 values may occur.
Time Series Examples
These are some time series plots from the dataset:
2006 BI3S-lab -
Hamburg, Germany - All rights reserved - Questions, Comments and
The Knowledge Portal on Forecasting with Neural Networks @ www.neural-forecasting.com - last update: 26.05.2009