In this demo, we will build an optimized fraud prediction model using EvalML. To optimize the pipeline, we will set up an objective function to minimize the percentage of total transaction value lost to fraud. At the end of this demo, we also show you how introducing the right objective during the training is over 4x better than using a generic machine learning metric like AUC.
[1]:
import evalml from evalml import AutoMLSearch from evalml.objectives import FraudCost
2020-08-06 20:07:57,525 featuretools - WARNING Featuretools failed to load plugin nlp_primitives from library nlp_primitives. For a full stack trace, set logging to debug.
/home/docs/checkouts/readthedocs.org/user_builds/feature-labs-inc-evalml/envs/v0.12.2/lib/python3.7/site-packages/evalml/pipelines/components/transformers/preprocessing/text_featurizer.py:31: RuntimeWarning: No text columns were given to TextFeaturizer, component will have no effect warnings.warn("No text columns were given to TextFeaturizer, component will have no effect", RuntimeWarning)
To optimize the pipelines toward the specific business needs of this model, you can set your own assumptions for the cost of fraud. These parameters are
retry_percentage - what percentage of customers will retry a transaction if it is declined?
retry_percentage
interchange_fee - how much of each successful transaction do you collect?
interchange_fee
fraud_payout_percentage - the percentage of fraud will you be unable to collect
fraud_payout_percentage
amount_col - the column in the data the represents the transaction amount
amount_col
Using these parameters, EvalML determines attempt to build a pipeline that will minimize the financial loss due to fraud.
[2]:
fraud_objective = FraudCost(retry_percentage=.5, interchange_fee=.02, fraud_payout_percentage=.75, amount_col='amount')
In order to validate the results of the pipeline creation and optimization process, we will save some of our data as a holdout set
[3]:
X, y = evalml.demos.load_fraud(n_rows=2500)
Number of Features Boolean 1 Categorical 6 Numeric 5 Number of training examples: 2500 Labels False 85.92% True 14.08% Name: fraud, dtype: object
EvalML natively supports one-hot encoding. Here we keep 1 out of the 6 categorical columns to decrease computation time.
[4]:
X = X.drop(['datetime', 'expiration_date', 'country', 'region', 'provider'], axis=1) X_train, X_holdout, y_train, y_holdout = evalml.preprocessing.split_data(X, y, test_size=0.2, random_state=0) print(X.dtypes)
card_id int64 store_id int64 amount int64 currency object customer_present bool lat float64 lng float64 dtype: object
Because the fraud labels are binary, we will use AutoMLSearch(problem_type='binary'). When we call .search(), the search for the best pipeline will begin.
AutoMLSearch(problem_type='binary')
.search()
[5]:
automl = AutoMLSearch(problem_type='binary', objective=fraud_objective, additional_objectives=['auc', 'f1', 'precision'], max_pipelines=5, optimize_thresholds=True) automl.search(X_train, y_train)
Generating pipelines to search over... ***************************** * Beginning pipeline search * ***************************** Optimizing for Fraud Cost. Lower score is better. Searching up to 5 pipelines. Allowed model families: random_forest, xgboost, catboost, linear_model
(1/5) Mode Baseline Binary Classification P... Elapsed:00:00 Starting cross validation Finished cross validation - mean Fraud Cost: 0.023 (2/5) CatBoost Classifier w/ Imputer Elapsed:00:00 Starting cross validation Finished cross validation - mean Fraud Cost: 0.002 (3/5) XGBoost Classifier w/ Imputer + One H... Elapsed:00:02 Starting cross validation Finished cross validation - mean Fraud Cost: 0.007 (4/5) Random Forest Classifier w/ Imputer +... Elapsed:01:00 Starting cross validation Finished cross validation - mean Fraud Cost: 0.002 (5/5) Logistic Regression Classifier w/ Imp... Elapsed:01:03 Starting cross validation Finished cross validation - mean Fraud Cost: 0.016 Search finished after 01:06 Best pipeline: CatBoost Classifier w/ Imputer Best pipeline Fraud Cost: 0.002316
Once the fitting process is done, we can see all of the pipelines that were searched, ranked by their score on the fraud detection objective we defined
[6]:
automl.rankings
to select the best pipeline we can run
[7]:
best_pipeline = automl.best_pipeline
You can get more details about any pipeline. Including how it performed on other objective functions.
[8]:
automl.describe_pipeline(automl.rankings.iloc[1]["id"])
********************************************************* * Random Forest Classifier w/ Imputer + One Hot Encoder * ********************************************************* Problem Type: Binary Classification Model Family: Random Forest Pipeline Steps ============== 1. Imputer * categorical_impute_strategy : most_frequent * numeric_impute_strategy : mean * fill_value : None 2. One Hot Encoder * top_n : 10 * categories : None * drop : None * handle_unknown : ignore * handle_missing : error 3. Random Forest Classifier * n_estimators : 100 * max_depth : 6 * n_jobs : -1 Training ======== Training for Binary Classification problems. Objective to optimize binary classification pipeline thresholds for: <evalml.objectives.fraud_cost.FraudCost object at 0x7f1c41c28e48> Total training time (including CV): 3.3 seconds Cross Validation ---------------- Fraud Cost AUC F1 Precision # Training # Testing 0 0.002 0.854 0.247 0.141 1066.000 667.000 1 0.002 0.849 0.247 0.141 1066.000 667.000 2 0.002 0.856 0.247 0.141 1067.000 666.000 mean 0.002 0.853 0.247 0.141 - - std 0.000 0.003 0.000 0.000 - - coef of var 0.055 0.004 0.001 0.001 - -
Finally, we retrain the best pipeline on all of the training data and evaluate on the holdout
[9]:
best_pipeline.fit(X_train, y_train)
<evalml.pipelines.utils.make_pipeline.<locals>.GeneratedPipeline at 0x7f1c3e34a240>
Now, we can score the pipeline on the hold out data using both the fraud cost score and the AUC.
[10]:
best_pipeline.score(X_holdout, y_holdout, objectives=["auc", fraud_objective])
OrderedDict([('AUC', 0.8225415282392027), ('Fraud Cost', 0.004329350526560073)])
To demonstrate the importance of optimizing for the right objective, let’s search for another pipeline using AUC, a common machine learning metric. After that, we will score the holdout data using the fraud cost objective to see how the best pipelines compare.
[11]:
automl_auc = AutoMLSearch(problem_type='binary', objective='auc', additional_objectives=['f1', 'precision'], max_pipelines=5, optimize_thresholds=True) automl_auc.search(X_train, y_train)
Generating pipelines to search over... ***************************** * Beginning pipeline search * ***************************** Optimizing for AUC. Greater score is better. Searching up to 5 pipelines. Allowed model families: random_forest, xgboost, catboost, linear_model
(1/5) Mode Baseline Binary Classification P... Elapsed:00:00 Starting cross validation Finished cross validation - mean AUC: 0.500 (2/5) CatBoost Classifier w/ Imputer Elapsed:00:00 Starting cross validation Finished cross validation - mean AUC: 0.844 (3/5) XGBoost Classifier w/ Imputer + One H... Elapsed:00:00 Starting cross validation Finished cross validation - mean AUC: 0.851 (4/5) Random Forest Classifier w/ Imputer +... Elapsed:01:05 Starting cross validation Finished cross validation - mean AUC: 0.856 (5/5) Logistic Regression Classifier w/ Imp... Elapsed:01:07 Starting cross validation Finished cross validation - mean AUC: 0.805 Search finished after 01:08 Best pipeline: Random Forest Classifier w/ Imputer + One Hot Encoder Best pipeline AUC: 0.856110
like before, we can look at the rankings and pick the best pipeline
[12]:
automl_auc.rankings
[13]:
best_pipeline_auc = automl_auc.best_pipeline # train on the full training data best_pipeline_auc.fit(X_train, y_train)
<evalml.pipelines.utils.make_pipeline.<locals>.GeneratedPipeline at 0x7f1c3db069e8>
[14]:
# get the fraud score on holdout data best_pipeline_auc.score(X_holdout, y_holdout, objectives=["auc", fraud_objective])
OrderedDict([('AUC', 0.8396345514950165), ('Fraud Cost', 0.004329350526560073)])
[15]:
# fraud score on fraud optimized again best_pipeline.score(X_holdout, y_holdout, objectives=["auc", fraud_objective])
When we optimize for AUC, we can see that the AUC score from this pipeline is better than the AUC score from the pipeline optimized for fraud cost. However, the losses due to fraud are over 3% of the total transaction amount when optimized for AUC and under 1% when optimized for fraud cost. As a result, we lose more than 2% of the total transaction amount by not optimizing for fraud cost specifically.
This happens because optimizing for AUC does not take into account the user-specified retry_percentage, interchange_fee, fraud_payout_percentage values. Thus, the best pipelines may produce the highest AUC but may not actually reduce the amount loss due to your specific type fraud.
This example highlights how performance in the real world can diverge greatly from machine learning metrics.