The explore package offers a simplified way to use machine learning to understand and explain patterns in the data.
explain_tree()
creates a decision tree. The target can
be binary, categorical or numericalexplain_forest()
creates a random forest. The target
can be binary, categorical or numericalexplain_logreg()
creates a logistic regression. The
target must be binarybalance_target()
to balance a targetweight_target()
to create weights for the decision
treeWe use synthetic data in this example
library(dplyr)
library(explore)
<- create_data_buy(obs = 1000)
data glimpse(data)
#> Rows: 1,000
#> Columns: 13
#> $ period <int> 202012, 202012, 202012, 202012, 202012, 202012, 202012…
#> $ buy <int> 1, 0, 1, 0, 0, 1, 0, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 1, …
#> $ age <int> 39, 57, 55, 66, 71, 44, 64, 51, 70, 44, 58, 47, 68, 71…
#> $ city_ind <int> 1, 0, 0, 0, 0, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, …
#> $ female_ind <int> 0, 0, 1, 0, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 1, 0, …
#> $ fixedvoice_ind <int> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, …
#> $ fixeddata_ind <int> 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, …
#> $ fixedtv_ind <int> 1, 0, 1, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 0, 1, 1, 1, 0, …
#> $ mobilevoice_ind <int> 0, 1, 1, 0, 0, 1, 0, 1, 1, 1, 0, 1, 0, 1, 1, 1, 0, 0, …
#> $ mobiledata_prd <chr> "NO", "NO", "MOBILE STICK", "NO", "BUSINESS", "BUSINES…
#> $ bbi_speed_ind <int> 1, 1, 0, 1, 0, 1, 0, 1, 0, 1, 1, 1, 0, 0, 1, 0, 0, 1, …
#> $ bbi_usg_gb <int> 77, 49, 53, 44, 55, 93, 50, 64, 63, 87, 45, 45, 70, 79…
#> $ hh_single <int> 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 1, 0, 0, 0, 0, 1, …
%>% explain_tree(target = buy) data
%>% explain_tree(target = mobiledata_prd) data
%>% explain_tree(target = age) data
%>% explain_forest(target = buy, ntree = 100) data
%>% explain_logreg(target = buy)
data #> # A tibble: 6 × 5
#> term estimate std.error statistic p.value
#> <chr> <dbl> <dbl> <dbl> <dbl>
#> 1 (Intercept) 5.87 0.544 10.8 3.88e-27
#> 2 age -0.146 0.0106 -13.8 3.49e-43
#> 3 city_ind 0.711 0.183 3.89 1.02e- 4
#> 4 female_ind 1.75 0.186 9.38 6.91e-21
#> 5 fixedtv_ind 1.51 0.190 7.93 2.14e-15
#> 6 bbi_usg_gb -0.0000724 0.0000904 -0.801 4.23e- 1
If you have a data set with a very unbalanced target (in this case
only 5% of all observations have buy == 1
) it may be
difficult to create a decision tree.
<- create_data_buy(obs = 2000, target1_prob = 0.05)
data %>% describe(buy)
data #> variable = buy
#> type = integer
#> na = 0 of 2 000 (0%)
#> unique = 2
#> 0 = 1 899 (95%)
#> 1 = 101 (5.1%)
It may help to balance the target before growing the decision tree
(or use weighs as alternative). In this example we down sample the data
so buy has 10% of target == 1
.
%>%
data balance_target(target = buy, min_prop = 0.10) %>%
explain_tree(target = buy)