Introduction to featurescreator • featurescreator

featurescreator is a package for creating features that capture important trends in data. It can be used in telecommunications, finance, or any other domain where time-based patterns in data are required.

For example, a telecommunications operator, wants to prevent churn by predicting which subscribers’ revenue will decline next month so that it can devise strategies to retain those customers. This type of prediction task necessitates the creation of a large number of features that capture trends in data over a period of several months for various dimensions such as recharge, revenue, number of complaints, network usage, number of active days, and so on. This package will assist them in developing features such as standard deviation for the previous n months, average for the previous n months, percentage change from m - 2 to m-1, and so on.

Import library

library(featurescreator)

Example data

example_data = tibble::tibble(
        subscriber_id = c(1, 2, 3),
        data_usage1 = c(10, 5, 3),  # 1 represent data usage in prediction month (m) - 1
        data_usage2 = c(4, 5, 6),  # m - 2
        data_usage3 = c(7, 8, 9),  # m - 3
        data_usage4 = c(10, 11, 12), # m - 4
        data_usage5 = c(13, 14, 15),  # m - 5
        othercolumn = c(5, 6, 7),  # Other example column
        data_usage_string6 = c(5, 6, 7)  # Other example column with an integer
    )

example_data
#> # A tibble: 3 × 8
#>   subscriber_id data_usage1 data_usage2 data_usage3 data_usage4 data_usage5
#>           <dbl>       <dbl>       <dbl>       <dbl>       <dbl>       <dbl>
#> 1             1          10           4           7          10          13
#> 2             2           5           5           8          11          14
#> 3             3           3           6           9          12          15
#> # … with 2 more variables: othercolumn <dbl>, data_usage_string6 <dbl>

get_matching_column_names

This function returns a list of matching columns given a column pattern.


featurescreator::get_matching_column_names(example_data, "data_usage")
#> [1] "data_usage1" "data_usage2" "data_usage3" "data_usage4" "data_usage5"

calculate_standard_deviation

This function computes the standard deviation of each subscriber’s data usage over all recorded months (rowwise). It takes two arguments: the dataframe to use and the pattern to match. The pattern to match is the prefix of the column name of interest not including the incrementing integer at the end.

featurescreator::calculate_standard_deviation(example_data, "data_usage")
#> [1] 2.792848 3.193744 3.872983

calculate_average

This function computes the average of each subscriber’s data usage over all recorded months (rowwise). It takes two arguments: the dataframe to use and the pattern to match. The pattern to match is the prefix of the column name of interest not including the incrementing integer at the end.

featurescreator::calculate_average(
  example_data,
  "data_usage"
)
#> [1] 8.8 8.6 9.0

calculate_percentage_change

The calculate_percentage_change function aims to generate new features to capture trends over time for a given comparison period. The compare_period argument is used to indicate which time periods to compare. For example, if we set it to (1, 1), the function calculates the percent change from the last month data usage (data_usage1), to the month before that (data_usage2).

featurescreator::calculate_percentage_change(example_data, "data_usage",
  compare_period = c(1, 1)
)
#> [1] 150   0 -50

Create percentage change of data usage in last 2 months (data_usage1, data_usage2) versus previous 2 months (data_usage3, data_usage4) by setting compare_period=(2, 2).

featurescreator::calculate_percentage_change(example_data, "data_usage",
  compare_period = c(2, 2)
)
#> [1] -17.64706 -47.36842 -57.14286

Create percentage change of data usage in last 2 months (data_usage1, data_usage2) versus previous 3 months (data_usage3, data_usage4, data_usage5) by setting compare_period=(2, 3). Comparison periods are brought to the same scales for a fare comparison.

featurescreator::calculate_percentage_change(example_data, "data_usage",
  compare_period = c(2, 3)
)
#> [1] -30.00000 -54.54545 -62.50000

To capture percentage changes for few particular months, a time_filter can be specified. Create percentage change of data usage in m - 2 and m - 3 (that is data_usage2 and data_usage3) versus m - 4 and m - 5 (that is data_usage4 and data_usage5) where m is the prediction month.

featurescreator::calculate_percentage_change(example_data, "data_usage",
  compare_period = c(2, 2),
  time_filter = c(2, 3, 4, 5)
)
#> [1] -52.17391 -48.00000 -44.44444