The evaluation metric for this competition is
[Mean Absolute Error](https://www.kaggle.com/wiki/MeanAbsoluteError)
. The mean absolute error is a common measure of forecast error in time series analysis and is sometimes known as "mean absolute deviation". Mean absolute error (MAE) is defined as.:
\[\frac{1}{n}\sum_{i=1}^n \left| y_i - \hat{y_i}\right| =\frac{1}{n}\sum_{i=1}^n \left| e_i \right|.\]
Where:
The error term is \[ |e_i| = |y_i-\hat{y_i}| \]
Actual units sold are \[y_i\]
Predicted units sold are \[\hat{y_i}\]
The number of forecasts made is \[n\]
Submission Format
Your submission file must have a header and should be structured in the following format. The id column is a triplet representing a store_nbr, item_nbr, and date. Form the id by concatenating these (in that order) with an underscore. E.g. "2_1_2013-06-15" represents store 2, item 1, sold on 2013-06-25. The units column is your submitted forecast. Your csv file should have one header row plus 120 forecast rows (2 stores * 2 products * 30 days = 120 forecasts).
id, units
1_1_2013-06-01,120
1_2_2013-06-01,110
2_1_2013-06-01,125
2_2_2013-06-01,204
etc.