Your task is to predict the probability of a client making a purchase in certain 8 categories in the next 7 days, so that the Bank can send relevant content (collections) to them.
The training (transactions_train.csv) and test (transactions_test.csv) datasets are identical and contain information on transactions of 25,000 unique clients each in the following format:
- сlient_dk – unique client number;
- trans_date – transaction date (represents the day number in chronological order, starting from the given date);
- small_group – a group of transactions characterizing the type of transaction (for example, grocery stores, clothing, gas stations, children's goods, etc.);
- amount – transaction amount in conventional units.
The quality of your recommendation system will be calculated based on \(averageROC_{AUC}\):
\[averageROC\ = \sum_{i=1}^8 {ROC_{AUCi} \over 8},\] where \(ROC_{AUCi}\) – is the average \(ROC_{AUC}\) for each of the 8 categories (i.e., for each category, the \(ROC_{AUC}\) for all Clients is first calculated, and then they are summed up and averaged by the number of categories).
Participants had to prepare a test.csv file with a structure identical to train_target.csv (25,000 Clients with client_dk and 8 categories - a total of 9 columns), filling in the purchase probabilities for each client in each of the above categories.
