| UID | DATE | PURCHASE_TOTAL |
|---|---|---|
| 63B454F18AC46B7B5A00DCCD | 2023-07-24 | 2800 |
| 63ED190F4C536A1AF7CF1263 | 2023-07-24 | 2800 |
| … | … | … |
| Metric | TYPE | DESCRIPTION |
|---|---|---|
| UID | String | Unique user ID |
| DATE | String | Date of purchase (YYYYY-MM-DD) |
| PURCHASE_TOTAL | Integer | Purchase value (Sum of all the items in the cart) |
| Requirement | |
|---|---|
| Record | Customer purchase |
| Size | > 10 000 recorda |
| Period | last 16 weeks |
| Data | The data includes users who have made at least one purchase in the past 16 weeks |
| Formats | .csv, separator , |
Before training the model, we perform data validation beforehand, but please make sure in advance that your data does not contain duplicates, missing values of required metrics, or no correspondence in the measurement of the same value (e.g. one record measures a purchase in USD and another in Cents).
The metrics specified in Template must be present in the data and correspond to the types specified in their descriptions. NULL values or its absence in any of the metrics will be considered invalid and the record will not be considered by the model.
Make sure your data covers and presents a complete picture of consumer behavior. For example, you may have uploaded data where you have missed purchase records of a popular brand. The model was trained on this data, but when the time comes to make a prediction the model encounters an unknown item or list of items in the user's shopping cart, it will be much more difficult for it to make an accurate prediction.
Make sure that your data is normally distributed over time. For example, 90% of records belong to the period of the last month and the remaining 10% belong to the other 3 months. Such distortion in the temporal data can greatly affect the accuracy of the model.