DataRobot API resources > API user guide > Common use cases > Python v2.x use cases > Anti-Money Laundering (AML) Alert Scoring > Anti-money laundering notebook

Anti-money laundering notebook¶

In this use case you will build a model that uses historical data, including customer and transactional information, to identify which alerts resulted in a Suspicious Activity Report (SAR). The model can then be used to assign a suspicious activity score to future alerts and improve the efficiency of an AML compliance program using rank ordering by score.

Download the sample training dataset here.

Import libraries¶

In [1]:

Copied!





!pip install datarobot --quiet

import datarobot as dr
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.ticker as mtick
import datetime as datetime
import os
import sys
from IPython.display import HTML
import json
import requests
import yaml

%matplotlib inline
light_blue = "#598fd6"
grey_blue = "#5f728b"
orange = "#dd6b3d"
!pip install datarobot --quiet

import datarobot as dr
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.ticker as mtick
import datetime as datetime
import os
import sys
from IPython.display import HTML
import json
import requests
import yaml

%matplotlib inline
light_blue = "#598fd6"
grey_blue = "#5f728b"
orange = "#dd6b3d"

Connect to DataRobot¶

In [2]:

Copied!

dr.Client()
# The `config_path` should only be specified if the config file is not in the default location described in the API Quickstart guide
# dr.Client(config_path = 'path-to-drconfig.yaml')
dr.Client()
# The `config_path` should only be specified if the config file is not in the default location described in the API Quickstart guide
# dr.Client(config_path = 'path-to-drconfig.yaml')

Out[2]:

<datarobot.rest.RESTClientObject at 0x7fb7a067f940>

Import data¶

In [2]:

Copied!





# Load the training dataset.
df = pd.read_csv("https://s3.amazonaws.com/datarobot-use-case-datasets/DR_Demo_AML_Alert_train.csv",
                  encoding = "ISO-8859-1")
df.info()
df.head()
# Load the training dataset.
df = pd.read_csv("https://s3.amazonaws.com/datarobot-use-case-datasets/DR_Demo_AML_Alert_train.csv",
                  encoding = "ISO-8859-1")
df.info()
df.head()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 10000 entries, 0 to 9999
Data columns (total 31 columns):
 #   Column                            Non-Null Count  Dtype  
---  ------                            --------------  -----  
 0   ALERT                             10000 non-null  int64  
 1   SAR                               10000 non-null  int64  
 2   kycRiskScore                      10000 non-null  int64  
 3   income                            9800 non-null   float64
 4   tenureMonths                      10000 non-null  int64  
 5   creditScore                       10000 non-null  int64  
 6   state                             10000 non-null  object 
 7   nbrPurchases90d                   10000 non-null  int64  
 8   avgTxnSize90d                     10000 non-null  float64
 9   totalSpend90d                     10000 non-null  float64
 10  csrNotes                          10000 non-null  object 
 11  nbrDistinctMerch90d               10000 non-null  int64  
 12  nbrMerchCredits90d                10000 non-null  int64  
 13  nbrMerchCreditsRndDollarAmt90d    10000 non-null  int64  
 14  totalMerchCred90d                 10000 non-null  float64
 15  nbrMerchCreditsWoOffsettingPurch  10000 non-null  int64  
 16  nbrPayments90d                    10000 non-null  int64  
 17  totalPaymentAmt90d                10000 non-null  float64
 18  overpaymentAmt90d                 10000 non-null  float64
 19  overpaymentInd90d                 10000 non-null  int64  
 20  nbrCustReqRefunds90d              10000 non-null  int64  
 21  indCustReqRefund90d               10000 non-null  int64  
 22  totalRefundsToCust90d             10000 non-null  float64
 23  nbrPaymentsCashLike90d            10000 non-null  int64  
 24  maxRevolveLine                    10000 non-null  int64  
 25  indOwnsHome                       10000 non-null  int64  
 26  nbrInquiries1y                    10000 non-null  int64  
 27  nbrCollections3y                  10000 non-null  int64  
 28  nbrWebLogins90d                   10000 non-null  int64  
 29  nbrPointRed90d                    10000 non-null  int64  
 30  PEP                               10000 non-null  int64  
dtypes: float64(7), int64(22), object(2)
memory usage: 2.4+ MB

Out[2]:

	ALERT	kycRiskScore	income	tenureMonths	creditScore	state	nbrPurchases90d	avgTxnSize90d	totalSpend90d	...	indCustReqRefund90d	totalRefundsToCust90d	nbrPaymentsCashLike90d	maxRevolveLine	indOwnsHome	nbrInquiries1y	nbrWebLogins90d	nbrPointRed90d
0	1	3	110300.0	5	757	PA	10	153.80	1538.00	...	1	45.82	5	6000	0	3	6	1
1	1	2	107800.0	6	715	NY	22	1.59	34.98	...	1	67.40	0	10000	1	3	87	0
2	1	1	74000.0	13	751	MA	7	57.64	403.48	...	1	450.69	0	10000	0	3	6	0
3	1	0	57700.0	1	659	NJ	14	29.52	413.28	...	1	71.43	0	8000	1	5	7	2
4	1	1	59800.0	3	709	PA	54	115.77	6251.58	...	1	2731.39	3	7000	1	1	8	1

5 rows × 31 columns

The sample data contains the following features:

ALERT: Alert Indicator
SAR: Target variable, SAR Indicator
kycRiskScore: Account relationship (Know Your Customer) score used at time of account opening
income: Annual income
tenureMonths: Account tenure in months
creditScore: Credit bureau score
state: Account billing address state
nbrPurchases90d: Number of purchases in last 90 days
avgTxnSize90d: Average transaction size in last 90 days
totalSpend90d: Total spend in last 90 days
csrNotes: Customer Service Representative notes and codes based on conversations with customer
nbrDistinctMerch90d: Number of distinct merchants purchased at in last 90 days
nbrMerchCredits90d: Number of credits from merchants in last 90 days
nbrMerchCreditsRndDollarAmt90d: Number of credits from merchants in round dollar amounts in last 90 days
totalMerchCred90d: Total merchant credit amount in last 90 days
nbrMerchCreditsWoOffsettingPurch: Number of merchant credits without an offsetting purchase in last 90 days
nbrPayments90d: Number of payments in last 90 days
totalPaymentAmt90d: Total payment amount in last 90 days
overpaymentAmt90d: Total amount overpaid in last 90 days
overpaymentInd90d: Indicator that account was overpaid in last 90 days
nbrCustReqRefunds90d: Number refund requests by the customer in last 90 days
indCustReqRefund90d: Indicator that customer requested a refund in last 90 days
totalRefundsToCust90d: Total refund amount in last 90 days
nbrPaymentsCashLike90d: Number of cash-like payments (e.g., money orders) in last 90 days
maxRevolveLine: Maximum revolving line of credit
indOwnsHome: Indicator that the customer owns a home
nbrInquiries1y: Number of credit inquiries in the past year
nbrCollections3y: Number of collections in the past year
nbrWebLogins90d: Number of logins to the bank website in the last 90 days
nbrPointRed90d: Number of loyalty point redemptions in the last 90 days
PEP: Politically Exposed Person indicator

Analyze, clean, and curate data¶

Preparing data is an iterative process. Even if you clean and prep your training data prior to uploading it, you can still improve its quality by performing Exploratory Data Analysis.

First, get a feel for the data. It is important to understand the dimensions of each feature and this provides an opportunity to ask questions if a feature doesn't make sense. Look at the distribution of the target variable and double check that the whole population only contains alerts (since this is the only population you care about).

In [3]:

Copied!





print("ALERT:")
print(df['ALERT'].value_counts(normalize=True))
print("SAR:")
print(df['SAR'].value_counts(normalize=True))
print("ALERT:")
print(df['ALERT'].value_counts(normalize=True))
print("SAR:")
print(df['SAR'].value_counts(normalize=True))

ALERT:
1    1.0
Name: ALERT, dtype: float64
SAR:
0    0.8974
1    0.1026
Name: SAR, dtype: float64

From the output can see that the false positive rate (SAR=0) is roughly 90%. This is expected for AML problems.

Next, assess the data quality by checking if any missing values are present in the data.

In [4]:

Copied!

df.isnull().sum()/len(df)
df.isnull().sum()/len(df)

Out[4]:

ALERT                               0.00
SAR                                 0.00
kycRiskScore                        0.00
income                              0.02
tenureMonths                        0.00
creditScore                         0.00
state                               0.00
nbrPurchases90d                     0.00
avgTxnSize90d                       0.00
totalSpend90d                       0.00
csrNotes                            0.00
nbrDistinctMerch90d                 0.00
nbrMerchCredits90d                  0.00
nbrMerchCreditsRndDollarAmt90d      0.00
totalMerchCred90d                   0.00
nbrMerchCreditsWoOffsettingPurch    0.00
nbrPayments90d                      0.00
totalPaymentAmt90d                  0.00
overpaymentAmt90d                   0.00
overpaymentInd90d                   0.00
nbrCustReqRefunds90d                0.00
indCustReqRefund90d                 0.00
totalRefundsToCust90d               0.00
nbrPaymentsCashLike90d              0.00
maxRevolveLine                      0.00
indOwnsHome                         0.00
nbrInquiries1y                      0.00
nbrCollections3y                    0.00
nbrWebLogins90d                     0.00
nbrPointRed90d                      0.00
PEP                                 0.00
dtype: float64

The data looks to be in good shape, with only 2% (which is very small) missing from the income variable. This is a time to question: is income missing because the person is unemployed or is this a system error?

Next, use the following code to visulize the data for better understanding. This is good practice that can help to spot anomalies or get additional information. This code selects numerical features (if there are categorical features, you can look at them later).

In [5]:

Copied!

plt_hist = df.select_dtypes(include='number').hist(figsize=(20, 20), xlabelsize=8, ylabelsize=8)
plt_hist = df.select_dtypes(include='number').hist(figsize=(20, 20), xlabelsize=8, ylabelsize=8)

Some quick observations based on the histograms:

Some features do not provide any useful information because they are all zeroes or have a single value (example: PEP).
Some features are zero-inflated (example: nbrPaymentsCashLike90d).
Some numerical features can be turned into categorical features (example: indOwnsHome)

Use the API to predict a SAR¶

Once you have configured your API credentials, endpoints, and environment, you can use the DataRobot API to do the following:

Upload a dataset.
Train a model to learn from the dataset on the Informative Features feature list.
Test prediction outcomes on the model with new data.
Deploy the model.
Predict outcomes on the deployed model using new data.

In [7]:

Copied!





# Upload a dataset
ct = datetime.datetime.now()
file_name = f"AML_Alert_train_{int(ct.timestamp())}.csv"

dataset = dr.Dataset.create_from_in_memory_data(df)
dataset.modify(name=file_name)

dataset
# Upload a dataset
ct = datetime.datetime.now()
file_name = f"AML_Alert_train_{int(ct.timestamp())}.csv"

dataset = dr.Dataset.create_from_in_memory_data(df)
dataset.modify(name=file_name)

dataset

Out[7]:

Dataset(name='AML_Alert_train_1658253378.csv', id='62d6f04257fa4f9a325372d5')

In [8]:

Copied!





# Create a new project based on dataset
ct = datetime.datetime.now()
project_name = f"Anti Money Laundering Alert Scoring_{int(ct.timestamp())}"

# The following steps use code to create the project and then use the DataRobot user interface (UI) to interpret results based on Leaderboard visualizations. 
# If you use the UI to create the project, choose `SAR` as the target.
project = dataset.create_project(project_name=project_name)
project_url = project.get_leaderboard_ui_permalink()

# Display the project ID and name. Your URL and project ID will be different.
print(f'''Project Details
Project URL: {project_url}
Project ID: {project.id}
Project Name: {project.project_name}
    ''')
# Create a new project based on dataset
ct = datetime.datetime.now()
project_name = f"Anti Money Laundering Alert Scoring_{int(ct.timestamp())}"

# The following steps use code to create the project and then use the DataRobot user interface (UI) to interpret results based on Leaderboard visualizations. 
# If you use the UI to create the project, choose `SAR` as the target.
project = dataset.create_project(project_name=project_name)
project_url = project.get_leaderboard_ui_permalink()

# Display the project ID and name. Your URL and project ID will be different.
print(f'''Project Details
Project URL: {project_url}
Project ID: {project.id}
Project Name: {project.project_name}
    ''')

Project Details
Project URL: https://app.datarobot.com/projects/62d6f07b5956dedb4c536e79/models
Project ID: 62d6f07b5956dedb4c536e79
Project Name: Anti Money Laundering Alert Scoring_1658253435

In [9]:

Copied!





# Feature lists control the subset of features that DataRobot uses to build models 
# You can use one of the automatically created lists, e.g. "Informative Features"
flists = project.get_featurelists()
flist = next(x for x in flists if x.name == "Informative Features")
flist.features
# Feature lists control the subset of features that DataRobot uses to build models 
# You can use one of the automatically created lists, e.g. "Informative Features"
flists = project.get_featurelists()
flist = next(x for x in flists if x.name == "Informative Features")
flist.features

Out[9]:

['SAR',
 'kycRiskScore',
 'income',
 'tenureMonths',
 'creditScore',
 'state',
 'nbrPurchases90d',
 'avgTxnSize90d',
 'totalSpend90d',
 'csrNotes',
 'nbrDistinctMerch90d',
 'nbrMerchCredits90d',
 'nbrMerchCreditsRndDollarAmt90d',
 'totalMerchCred90d',
 'nbrMerchCreditsWoOffsettingPurch',
 'nbrPayments90d',
 'totalPaymentAmt90d',
 'overpaymentAmt90d',
 'overpaymentInd90d',
 'nbrCustReqRefunds90d',
 'totalRefundsToCust90d',
 'nbrPaymentsCashLike90d',
 'maxRevolveLine',
 'indOwnsHome',
 'nbrInquiries1y',
 'nbrCollections3y',
 'nbrWebLogins90d',
 'nbrPointRed90d']

In [10]:

Copied!





# Select modeling parameters and start the modeling process
project.set_target(target='SAR',
                   mode=dr.AUTOPILOT_MODE.QUICK,
                   featurelist_id=flist.id,
                   worker_count='-1')

display(HTML(f'For a full reference of available parameters, see <a target="_blank" rel="noopener noreferrer" href="https://datarobot-public-api-client.readthedocs-hosted.com/page/autodoc/api_reference.html#datarobot.models.Project.set_target">Project.set_target.</a>'))

project.wait_for_autopilot(check_interval=20.0, timeout=86400, verbosity=0)
# Select modeling parameters and start the modeling process
project.set_target(target='SAR',
                   mode=dr.AUTOPILOT_MODE.QUICK,
                   featurelist_id=flist.id,
                   worker_count='-1')

display(HTML(f'For a full reference of available parameters, see Project.set_target.'))

project.wait_for_autopilot(check_interval=20.0, timeout=86400, verbosity=0)

For a full reference of available parameters, see Project.set_target.

Models in the UI¶

Once the project is started, you can start exploring in the UI—even while Autopilot is running and building models. You can open the project from the Manage Projects center. Or, retrieve the project URL with the snippet below and use the output to navigate to the DataRobot application.

In [11]:

Copied!

# Display project ID and URL
project_url
# Display project ID and URL
project_url

Out[11]:

'https://app.datarobot.com/projects/62d6f07b5956dedb4c536e79/models'

Exploratory Data Analysis (EDA)¶

Navigate to the Data tab to learn more about your data.

Click each feature to see a variety of information, including a histogram that represents the relationship of the feature with the target.