Abnormal transaction detection
Karim Mrezhoud
2019-10-06
- 0.0.1 Load packages
- 0.0.2 read data
- 0.0.3 Glimpse for data quality
- 0.0.4 Join transactions and identities dataframes
- 1 EDA
- 2 Modeling with LighGBM
- 3 Prediction and Submission
- 4 Freatures Engineering
0.0.1 Load packages
library(rmarkdown)
library(knitr)
library(readr)
library(tidyverse)
library(data.table)
library(MLmetrics)
library(lightgbm)
library(lubridate)
library(plyr)
library(moments)
library(dplyr)
library(rattle)
library(rpart)
library(tictoc)
library(inspectdf)
options(warn=-1)
options(scipen = 99)
#os.environ['KMP_DUPLICATE_LIB_OK']='True'0.0.2 read data
## Error in setwd("/media/kirus/DATA/learn_by_example/Fraud_detection/IEEE-CIS"): cannot change working directorylocal <- TRUE
if(local == TRUE){
train_iden <- fread("dataset/train_identity.csv")
train_trans <- fread("dataset/train_transaction.csv")
test_iden <- fread("dataset/test_identity.csv")
test_trans <- fread("dataset/test_transaction.csv")
} else{
train_iden <- fread("../input/train_identity.csv")
train_trans <- fread("../input/train_transaction.csv")
test_iden <- fread("../input/test_identity.csv")
test_trans <- fread("../input/test_transaction.csv")
}0.0.3 Glimpse for data quality
The training data-set seems to have a lot of missing values.
0.0.4 Join transactions and identities dataframes
y_train <- train_trans$isFraud
train_trans$isFraud <- NULL
train <- train_trans %>% left_join(train_iden)## Joining, by = "TransactionID"## Joining, by = "TransactionID"1 EDA
1.1 Explore missing variables
missing_train <- colSums(is.na(train))[colSums(is.na(train)) > 0] %>% sort(decreasing=TRUE)
missing_test <- colSums(is.na(test))[colSums(is.na(test)) > 0] %>% sort(decreasing=TRUE)
head(missing_test)## id_24 id_25 id_26 id_07 id_08 id_21
## 501951 501652 501644 501632 501632 5016321.1.1 Ratio of missing variables
# Ratio of missing values
missing_train_pct <- round(missing_train/nrow(train), 2)
missing_test_pct <- round(missing_test/nrow(test), 2)
# drop variable with more than 0.75 of missing values
drop_col_train <- names(missing_train_pct[missing_train_pct > 0.75])
drop_col_test <- names(missing_test_pct[missing_test_pct > 0.75])
all(drop_col_test %in% drop_col_train) # TRUE, it means all drop_col_test are in drop_col_train## [1] TRUE## [1] "D6" "id_13" "V217" "V218" "V219"
## [6] "V223" "V224" "V225" "V226" "V228"
## [11] "V229" "V230" "V231" "V232" "V233"
## [16] "V235" "V236" "V237" "V240" "V241"
## [21] "V242" "V243" "V244" "V246" "V247"
## [26] "V248" "V249" "V252" "V253" "V254"
## [31] "V257" "V258" "V260" "V261" "V262"
## [36] "V263" "V264" "V265" "V266" "V267"
## [41] "V268" "V269" "V273" "V274" "V275"
## [46] "V276" "V277" "V278" "id_05" "id_06"
## [51] "id_20" "id_19" "id_17" "V167" "V168"
## [56] "V172" "V173" "V176" "V177" "V178"
## [61] "V179" "V181" "V182" "V183" "V186"
## [66] "V187" "V190" "V191" "V192" "V193"
## [71] "V196" "V199" "V202" "V203" "V204"
## [76] "V205" "V206" "V207" "V211" "V212"
## [81] "V213" "V214" "V215" "V216" "V169"
## [86] "V170" "V171" "V174" "V175" "V180"
## [91] "V184" "V185" "V188" "V189" "V194"
## [96] "V195" "V197" "V198" "V200" "V201"
## [101] "V208" "V209" "V210" "id_02" "id_11"
## [106] "V220" "V221" "V222" "V227" "V234"
## [111] "V238" "V239" "V245" "V250" "V251"
## [116] "V255" "V256" "V259" "V270" "V271"
## [121] "V272" "id_01" "id_12" "id_15" "id_16"
## [126] "id_23" "id_27" "id_28" "id_29" "id_30"
## [131] "id_31" "id_33" "id_34" "id_35" "id_36"
## [136] "id_37" "id_38" "DeviceType" "DeviceInfo"There is more dropped variables in train dataset that test dataset if we use a rate of 0.75 missed values. All theses listed variables do not have more than rates bigger than 0.75 in test dataset.
1.1.2 Set missing rate to drop the same variables in train and test datasets
# drop variable with more than 0.75 of missing values
drop_col_train <- names(missing_train_pct[missing_train_pct > 0.7])
drop_col_test <- names(missing_test_pct[missing_test_pct > 0.7])
all(drop_col_test %in% drop_col_train) # TRUE, it means all drop_col_test are in drop_col_train## [1] TRUE## character(0)## character(0)When we used a threshold of 0.7, we obtain the same list of variables that we can drop for tarin and test datasets.
1.2 Potential variables to drop
## [1] 207## [1] "id_24" "id_25" "id_07" "id_08" "id_21"
## [6] "id_26" "id_22" "dist2" "D7" "id_18"
## [11] "D13" "D14" "D12" "id_03" "id_04"
## [16] "D6" "D8" "D9" "id_09" "id_10"
## [21] "id_32" "id_14" "V138" "V139" "V140"
## [26] "V141" "V142" "V146" "V147" "V148"
## [31] "V149" "V153" "V154" "V155" "V156"
## [36] "V157" "V158" "V161" "V162" "V163"
## [41] "V143" "V144" "V145" "V150" "V151"
## [46] "V152" "V159" "V160" "V164" "V165"
## [51] "V166" "V322" "V323" "V324" "V325"
## [56] "V326" "V327" "V328" "V329" "V330"
## [61] "V331" "V332" "V333" "V334" "V335"
## [66] "V336" "V337" "V338" "V339" "id_13"
## [71] "V217" "V218" "V219" "V223" "V224"
## [76] "V225" "V226" "V228" "V229" "V230"
## [81] "V231" "V232" "V233" "V235" "V236"
## [86] "V237" "V240" "V241" "V242" "V243"
## [91] "V244" "V246" "V247" "V248" "V249"
## [96] "V252" "V253" "V254" "V257" "V258"
## [101] "V260" "V261" "V262" "V263" "V264"
## [106] "V265" "V266" "V267" "V268" "V269"
## [111] "V273" "V274" "V275" "V276" "V277"
## [116] "V278" "id_05" "id_06" "id_20" "id_19"
## [121] "id_17" "V167" "V168" "V172" "V173"
## [126] "V176" "V177" "V178" "V179" "V181"
## [131] "V182" "V183" "V186" "V187" "V190"
## [136] "V191" "V192" "V193" "V196" "V199"
## [141] "V202" "V203" "V204" "V205" "V206"
## [146] "V207" "V211" "V212" "V213" "V214"
## [151] "V215" "V216" "V169" "V170" "V171"
## [156] "V174" "V175" "V180" "V184" "V185"
## [161] "V188" "V189" "V194" "V195" "V197"
## [166] "V198" "V200" "V201" "V208" "V209"
## [171] "V210" "id_02" "id_11" "V220" "V221"
## [176] "V222" "V227" "V234" "V238" "V239"
## [181] "V245" "V250" "V251" "V255" "V256"
## [186] "V259" "V270" "V271" "V272" "id_01"
## [191] "id_12" "id_15" "id_16" "id_23" "id_27"
## [196] "id_28" "id_29" "id_30" "id_31" "id_33"
## [201] "id_34" "id_35" "id_36" "id_37" "id_38"
## [206] "DeviceType" "DeviceInfo"We can optimize missing value rate threshold by training and validation scores.
1.3 Merge train and test dataset
1.4 Inspect the type of variables
There are 399 numeric variables, 32 categorical variables and 3 integer variables. We can reduce memory if we convert numeric variables to integer.
#numeric_vars_stat <- inspect_num(full)
#numeric_vars <- numeric_vars_stat$col_name
#numeric_vars_stat
numeric_vars <- names(full)[sapply(full, class) == "numeric"]1.4.1 Convert numeric variables to integer (reduce computing memory)
library(tictoc)
# If values are integer type, value==floor(value)
is_int <- function(x){
fnum <- fivenum(x)
return(identical(fnum, floor(fnum)))
}
tic("check is integer")
int_idx <- sapply(full[numeric_vars], is_int)
toc()## check is integer: 36.963 sec elapsed#identical(fivenum(full$card2), floor(full$card2))
#is_int(full$card2)
int_vars <- names(int_idx)[int_idx]
paste("Number of numeric variables that we can convert to interger:", length(int_vars))## [1] "Number of numeric variables that we can convert to interger: 372"## [1] "Before : 3620.8 Mb"full[int_vars] <- lapply(full[int_vars], as.integer)
after <- object.size(full)
print(paste("After :", format(after, units = "MB") )) ## [1] "After : 2063.8 Mb"
1.4.2 Categorical variables
categorical_vars_stat <- inspect_cat(full)
categorical_vars <- categorical_vars_stat$col_name
categorical_vars_stat ## # A tibble: 32 x 5
## col_name cnt common common_pcnt levels
## <chr> <int> <chr> <dbl> <list>
## 1 card4 5 visa 65.6 <tibble [5 × 3]>
## 2 card6 5 debit 75.2 <tibble [5 × 3]>
## 3 DeviceInfo 2801 <NA> 73.9 <tibble [2,801 × 3]>
## 4 DeviceType 4 <NA> 73.9 <tibble [4 × 3]>
## 5 id_12 3 <NA> 73.9 <tibble [3 × 3]>
## 6 id_15 5 <NA> 73.9 <tibble [5 × 3]>
## 7 id_16 4 <NA> 73.9 <tibble [4 × 3]>
## 8 id_23 5 <NA> 73.9 <tibble [5 × 3]>
## 9 id_27 4 <NA> 73.9 <tibble [4 × 3]>
## 10 id_28 4 <NA> 73.9 <tibble [4 × 3]>
## # … with 22 more rows1.4.3 Transform categorical variables to factor
1.4.4 Label encoding of factors
##
## 1 2
## 506691 590540
2 Modeling with LighGBM
2.1 remove variables with high rate of missing values (0.7)
data.table::fwrite(full,file = "full.csv")
#full <- fread(file = "full.csv")
full_clean <- full %>%
select(-drop_col_train)
rm(full)
invisible(gc())
dim(full_clean)## [1] 1097231 2272.2 Preprocessing
X_train <- full_clean[full_clean$key=='2', ] %>% select( -TransactionID)
length(y_train ) == dim(X_train)[1]## [1] TRUE2.3 relace NA by blank
LightGBM support by default NA’s values. We can desable this option by use_missing=false.
# Replace NA's by blank
X_train[is.na(X_train)] <- ""
colSums(is.na(X_train))[colSums(is.na(X_train)) > 0] %>% sort(decreasing=TRUE)## named numeric(0)2.4 Light GBM parameters
lgb_param <- list(
boosting_type = 'dart',
objective = "binary" ,
metric = "AUC",
boost_from_average = "false",
tree_learner = "serial",
max_depth = -1,
learning_rate = 0.01,
num_leaves = 197,
feature_fraction = 0.3,
bagging_freq = 1,
bagging_fraction = 0.7,
min_data_in_leaf = 100,
bagging_seed = 11,
max_bin = 255)2.5 Split train dataset for validation
#set.seed(35)
#train.idx <- sample(nrow(X_train), 0.75*nrow(X_train))
# dtrain <- lgb.Dataset(data=as.matrix(X_train[train.idx,]), label=y_train[train.idx], free_raw_data=FALSE)
#
# dvalid <- lgb.Dataset(data=as.matrix(X_train[-train.idx, ]), label=y_train[-train.idx], free_raw_data=FALSE)
#
# invisible(gc())2.6 Train the model
2.7 Best Iteration
2.8 Training with full train dataset
2.9 ## Save model
3 Prediction and Submission
4 Freatures Engineering
4.0.1 Set python version and anaconda environment
## python: /Users/Mezhoud/venv/bin/python3
## libpython: /Users/Mezhoud/miniconda3/lib/libpython3.7m.dylib
## pythonhome: /Users/Mezhoud/miniconda3:/Users/Mezhoud/miniconda3
## virtualenv: /Users/Mezhoud/venv/bin/activate_this.py
## version: 3.7.1 (default, Dec 14 2018, 13:28:58) [Clang 4.0.1 (tags/RELEASE_401/final)]
## numpy: /Users/Mezhoud/miniconda3/lib/python3.7/site-packages/numpy
## numpy_version: 1.16.4
##
## NOTE: Python version was forced by use_python function4.0.2 Reload raw dataset
train_iden <- fread("dataset/train_identity.csv")
train_trans <- fread("dataset/train_transaction.csv")
train <- train_trans %>% left_join(train_iden)## Joining, by = "TransactionID"4.1 Working with sample dataset
## card1 card2 card3 card4 card5 card6
## 1 13926 NA 150 discover 142 credit
## 2 2755 404 150 mastercard 102 credit
## 3 4663 490 150 visa 166 debit
## 4 18132 567 150 mastercard 117 debit
## 5 4497 514 150 mastercard 102 credit
## 6 5937 555 150 visa 226 debit## id_01 id_02 id_03 id_04 id_05 id_06 id_07 id_08 id_09 id_10 id_11
## 1 NA NA NA NA NA NA NA NA NA NA NA
## 2 NA NA NA NA NA NA NA NA NA NA NA
## 3 NA NA NA NA NA NA NA NA NA NA NA
## 4 NA NA NA NA NA NA NA NA NA NA NA
## 5 0 70787 NA NA NA NA NA NA NA NA 100
## 6 NA NA NA NA NA NA NA NA NA NA NA
## id_12 id_13 id_14 id_15 id_16 id_17 id_18 id_19 id_20 id_21 id_22
## 1 <NA> NA NA <NA> <NA> NA NA NA NA NA NA
## 2 <NA> NA NA <NA> <NA> NA NA NA NA NA NA
## 3 <NA> NA NA <NA> <NA> NA NA NA NA NA NA
## 4 <NA> NA NA <NA> <NA> NA NA NA NA NA NA
## 5 NotFound NA -480 New NotFound 166 NA 542 144 NA NA
## 6 <NA> NA NA <NA> <NA> NA NA NA NA NA NA
## id_23 id_24 id_25 id_26 id_27 id_28 id_29 id_30
## 1 <NA> NA NA NA <NA> <NA> <NA> <NA>
## 2 <NA> NA NA NA <NA> <NA> <NA> <NA>
## 3 <NA> NA NA NA <NA> <NA> <NA> <NA>
## 4 <NA> NA NA NA <NA> <NA> <NA> <NA>
## 5 NA NA NA New NotFound Android 7.0
## 6 <NA> NA NA NA <NA> <NA> <NA> <NA>
## id_31 id_32 id_33 id_34 id_35 id_36 id_37
## 1 <NA> NA <NA> <NA> <NA> <NA> <NA>
## 2 <NA> NA <NA> <NA> <NA> <NA> <NA>
## 3 <NA> NA <NA> <NA> <NA> <NA> <NA>
## 4 <NA> NA <NA> <NA> <NA> <NA> <NA>
## 5 samsung browser 6.2 32 2220x1080 match_status:2 T F T
## 6 <NA> NA <NA> <NA> <NA> <NA> <NA>
## id_38 DeviceType DeviceInfo
## 1 <NA> <NA> <NA>
## 2 <NA> <NA> <NA>
## 3 <NA> <NA> <NA>
## 4 <NA> <NA> <NA>
## 5 T mobile SAMSUNG SM-G892A Build/NRD90M
## 6 <NA> <NA> <NA>## Observations: 5,000
## Variables: 6
## $ card1 <int> 13926, 2755, 4663, 18132, 4497, 5937, 12308, 12695, 2803, …
## $ card2 <dbl> NA, 404, 490, 567, 514, 555, 360, 490, 100, 111, 352, 375,…
## $ card3 <dbl> 150, 150, 150, 150, 150, 150, 150, 150, 150, 150, 117, 185…
## $ card4 <chr> "discover", "mastercard", "visa", "mastercard", "mastercar…
## $ card5 <dbl> 142, 102, 166, 117, 102, 226, 166, 226, 226, 224, 134, 224…
## $ card6 <chr> "credit", "credit", "debit", "debit", "credit", "debit", "…## Observations: 5,000
## Variables: 40
## $ id_01 <dbl> NA, NA, NA, NA, 0, NA, NA, NA, -5, NA, -5, -5, NA, NA…
## $ id_02 <dbl> NA, NA, NA, NA, 70787, NA, NA, NA, 98945, NA, 191631,…
## $ id_03 <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, 0, NA, NA, NA…
## $ id_04 <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, 0, NA, NA, NA…
## $ id_05 <dbl> NA, NA, NA, NA, NA, NA, NA, NA, 0, NA, 0, 0, NA, NA, …
## $ id_06 <dbl> NA, NA, NA, NA, NA, NA, NA, NA, -5, NA, 0, -6, NA, NA…
## $ id_07 <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, N…
## $ id_08 <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, N…
## $ id_09 <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, 0, NA, NA, NA…
## $ id_10 <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, 0, NA, NA, NA…
## $ id_11 <dbl> NA, NA, NA, NA, 100, NA, NA, NA, 100, NA, 100, 100, N…
## $ id_12 <chr> NA, NA, NA, NA, "NotFound", NA, NA, NA, "NotFound", N…
## $ id_13 <dbl> NA, NA, NA, NA, NA, NA, NA, NA, 49, NA, 52, 52, NA, N…
## $ id_14 <dbl> NA, NA, NA, NA, -480, NA, NA, NA, -300, NA, NA, NA, N…
## $ id_15 <chr> NA, NA, NA, NA, "New", NA, NA, NA, "New", NA, "Found"…
## $ id_16 <chr> NA, NA, NA, NA, "NotFound", NA, NA, NA, "NotFound", N…
## $ id_17 <dbl> NA, NA, NA, NA, 166, NA, NA, NA, 166, NA, 121, 225, N…
## $ id_18 <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, N…
## $ id_19 <dbl> NA, NA, NA, NA, 542, NA, NA, NA, 621, NA, 410, 176, N…
## $ id_20 <dbl> NA, NA, NA, NA, 144, NA, NA, NA, 500, NA, 142, 507, N…
## $ id_21 <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, N…
## $ id_22 <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, N…
## $ id_23 <chr> NA, NA, NA, NA, "", NA, NA, NA, "", NA, "", "", NA, N…
## $ id_24 <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, N…
## $ id_25 <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, N…
## $ id_26 <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, N…
## $ id_27 <chr> NA, NA, NA, NA, "", NA, NA, NA, "", NA, "", "", NA, N…
## $ id_28 <chr> NA, NA, NA, NA, "New", NA, NA, NA, "New", NA, "Found"…
## $ id_29 <chr> NA, NA, NA, NA, "NotFound", NA, NA, NA, "NotFound", N…
## $ id_30 <chr> NA, NA, NA, NA, "Android 7.0", NA, NA, NA, "iOS 11.1.…
## $ id_31 <chr> NA, NA, NA, NA, "samsung browser 6.2", NA, NA, NA, "m…
## $ id_32 <dbl> NA, NA, NA, NA, 32, NA, NA, NA, 32, NA, NA, NA, NA, N…
## $ id_33 <chr> NA, NA, NA, NA, "2220x1080", NA, NA, NA, "1334x750", …
## $ id_34 <chr> NA, NA, NA, NA, "match_status:2", NA, NA, NA, "match_…
## $ id_35 <chr> NA, NA, NA, NA, "T", NA, NA, NA, "T", NA, "F", "F", N…
## $ id_36 <chr> NA, NA, NA, NA, "F", NA, NA, NA, "F", NA, "F", "F", N…
## $ id_37 <chr> NA, NA, NA, NA, "T", NA, NA, NA, "F", NA, "T", "T", N…
## $ id_38 <chr> NA, NA, NA, NA, "T", NA, NA, NA, "T", NA, "T", "T", N…
## $ DeviceType <chr> NA, NA, NA, NA, "mobile", NA, NA, NA, "mobile", NA, "…
## $ DeviceInfo <chr> NA, NA, NA, NA, "SAMSUNG SM-G892A Build/NRD90M", NA, …4.1.1 Inspect missing value
## # A tibble: 6 x 3
## col_name cnt pcnt
## <chr> <int> <dbl>
## 1 card2 75 1.5
## 2 card5 9 0.18
## 3 card1 0 0
## 4 card3 0 0
## 5 card4 0 0
## 6 card6 0 0## # A tibble: 40 x 3
## col_name cnt pcnt
## <chr> <int> <dbl>
## 1 id_24 4968 99.4
## 2 id_07 4967 99.3
## 3 id_08 4967 99.3
## 4 id_21 4967 99.3
## 5 id_22 4967 99.3
## 6 id_25 4967 99.3
## 7 id_26 4967 99.3
## 8 id_18 4718 94.4
## 9 id_03 4578 91.6
## 10 id_04 4578 91.6
## # … with 30 more rows## card1 card2 card3 card4
## Min. : 1033 Min. :100.0 Min. :100.0 Length:5000
## 1st Qu.: 6459 1st Qu.:194.0 1st Qu.:150.0 Class :character
## Median : 9335 Median :321.0 Median :150.0 Mode :character
## Mean : 9737 Mean :346.5 Mean :152.3
## 3rd Qu.:13481 3rd Qu.:500.0 3rd Qu.:150.0
## Max. :18390 Max. :600.0 Max. :191.0
## NA's :75
## card5 card6
## Min. :100 Length:5000
## 1st Qu.:166 Class :character
## Median :224 Mode :character
## Mean :200
## 3rd Qu.:226
## Max. :237
## NA's :9Card1, 4 and 6 do not have a missing values. Card 4 and 6 are character and card 1 is numeric.
4.1.2 Replace NA by 0
## card1 card2 card3 card4
## Min. : 1033 Min. : 0.0 Min. :100.0 Length:5000
## 1st Qu.: 6459 1st Qu.:194.0 1st Qu.:150.0 Class :character
## Median : 9335 Median :321.0 Median :150.0 Mode :character
## Mean : 9737 Mean :341.3 Mean :152.3
## 3rd Qu.:13481 3rd Qu.:494.0 3rd Qu.:150.0
## Max. :18390 Max. :600.0 Max. :191.0
## card5 card6
## Min. : 0.0 Length:5000
## 1st Qu.:166.0 Class :character
## Median :224.0 Mode :character
## Mean :199.7
## 3rd Qu.:226.0
## Max. :237.04.2 Encrypt cards and Devices with sha hash functions
4.3 Groupe by hashes card and devices for isFraud transaction
4.3.1 Most device used for Frauds (43 transactions)
train_5000 %>%
select(card_hash, device_hash, isFraud) %>%
group_by( card_hash, device_hash, isFraud) %>%
add_tally() %>%
filter(isFraud == 1) %>%
distinct() %>%
count("device_hash") %>%
arrange(desc(freq))## device_hash freq
## 1 b0eec9231ae82c7 43
## 2 b40c174a01697ba 8
## 3 1bd5a7c38756eb4 4
## 4 d524c1a0811da49 4
## 5 ad0380c681c6f29 3
## 6 68693d02ab4fbb2 2
## 7 8645585e38a597c 2
## 8 d1ed60daba157ae 2
## 9 a5580d69ef89536 1
## 10 c3a1e3691b7af20 1
## 11 d2ad3adab1c84ef 14.3.2 Most cards used for Frauds
train_5000 %>%
select(card_hash, device_hash, isFraud) %>%
group_by( card_hash, device_hash, isFraud) %>%
add_tally() %>%
filter(isFraud == 1) %>%
distinct() %>%
count("card_hash") %>%
arrange(desc(freq))## card_hash freq
## 1 73d559ba355793c 2
## 2 ba335500048094d 2
## 3 c8e2247b6b1dd70 2
## 4 d33188e37e3cbde 2
## 5 e5218c4a6ba6b0d 2
## 6 02089e4201c478a 1
## 7 046089c4e6ece42 1
## 8 04880baba35b945 1
## 9 09a90b8bd21f5f0 1
## 10 0cac5612d2f3305 1
## 11 12e54d765cf4f88 1
## 12 179c088a27a4312 1
## 13 18122e9b2c8a47f 1
## 14 196fc3b72f0cf45 1
## 15 270fe39b31404fb 1
## 16 2911521a2650174 1
## 17 2963f1130483f86 1
## 18 2db26f3fccaeb06 1
## 19 354027dc5af1138 1
## 20 355b405219be922 1
## 21 3dcac3668b3812d 1
## 22 41e70506bdbc666 1
## 23 448155c2c23e3d2 1
## 24 48e958b429f3a7c 1
## 25 4d7e2f6c08a5805 1
## 26 53be1a94a5a5750 1
## 27 543212993b885f8 1
## 28 56210b84c6eeb06 1
## 29 5ab407639145b99 1
## 30 697e04bfea34575 1
## 31 6f473e021469c90 1
## 32 74a6fdb87df1bf9 1
## 33 7f6cbfe70bade06 1
## 34 7f985ed6bafd564 1
## 35 81e55ba2fb9c7f5 1
## 36 82da7cb095aad12 1
## 37 832c88c03f01511 1
## 38 8c1479ec0bda456 1
## 39 9a0e5a6e6dd9a21 1
## 40 9a461887c830051 1
## 41 9de338acc31c454 1
## 42 a0b81fd3bc73509 1
## 43 a2271484fc98d37 1
## 44 a46fd032bdebece 1
## 45 a52b14cd71a5329 1
## 46 a7ef1b6a8a0258e 1
## 47 b3926b2c86a782f 1
## 48 b72993e4926ae45 1
## 49 bbdccfb754b3c0c 1
## 50 be9b8946d6b309a 1
## 51 c3a98cc10558651 1
## 52 c8138c8d90570bf 1
## 53 ccd711d9a56bbb9 1
## 54 d2c3243d549807e 1
## 55 d2ca229767cb03d 1
## 56 d90a14c4d36ab75 1
## 57 dbb434897c459e9 1
## 58 dd6c6e6c88e4d79 1
## 59 e3bc3cc98381725 1
## 60 e7cb958bc9fcdbe 1
## 61 e9c8b964d4806b3 1
## 62 f137006a8f5170a 1
## 63 f2c3c2de908a477 1
## 64 f8360028f0c401d 1
## 65 f9f318efd5e4511 1
## 66 fe826cddbef2752 1