Preprocessing

The preprocess function transforms a transactional table into a customer aggregated table with custom options for aggregation methods for numeric and categorical columns.

In the current state we assume that the input data is a transaction table with 4 columns required (transactionid, id, orderdate, transactionvalue). The function will raise an error if these column names are not found.

formatted <- preprocess(citrus::transactional_data)

Example 1: numeric columns

If nothing other than numeric_operation_list = NA, the preprocessing will default to RFM preprocessing if no categorical columns exist in the data set. It can be used in e.g. unsupervised learning.

formatted <- preprocess(citrus::transactional_data, numeric_operation_list = NA)

The numeric columns can be aggregated using various aggregation methods like e.g. ‘min’ and standard deviation ‘sd’.

formatted <- preprocess(citrus::transactional_data, 
                        numeric_operation_list = c('min', 'sd')
                        )

Example 2: categorical columns

When the table contains categorical columns, the preprocess function check if any categorical column exists and assign the most common category for each column to each user.

formatted <- preprocess(citrus::transactional_data, 
                        target = "transactionvalue"
                        )
#> Calculating target values

It is possible to pass a list of categorical columns to include in the final table. In this example the function uses the most common category for each user in the column ‘country’.

formatted <- preprocess(citrus::transactional_data, 
                        categories = c("country"), 
                        target = "transactionvalue"
                        )
#> Calculating target values

Example 3: target column

Specify the target variable and the aggregation function to perform on it if you want to run a supervised model. In the example the target column is the mean of the transactionvalue column.

formatted <- preprocess(citrus::transactional_data,
                        categories = c('country'), 
                        numeric_operation_list = c('min', 'sd'),
                        target = 'transactionvalue', 
                        target_agg = 'mean')
#> Calculating target values