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
To optimize the pipelines toward the specific business needs of this model, we can set our 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 the holdout set.
[3]:
X, y = evalml.demos.load_fraud(n_rows=1000)
Number of Features Boolean 1 Categorical 6 Numeric 5 Number of training examples: 1000 Targets False 85.90% True 14.10% 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]:
cols_to_drop = ['datetime', 'expiration_date', 'country', 'region', 'provider'] for col in cols_to_drop: X.pop(col) X_train, X_holdout, y_train, y_holdout = evalml.preprocessing.split_data(X, y, problem_type='binary', test_size=0.2, random_state=0) print(X.types)
Physical Type Logical Type Semantic Tag(s) Data Column card_id Int64 Integer ['numeric'] store_id Int64 Integer ['numeric'] amount Int64 Integer ['numeric'] currency category Categorical ['category'] customer_present boolean Boolean [] lat float64 Double ['numeric'] lng float64 Double ['numeric']
Because the fraud labels are binary, we will use AutoMLSearch(X_train=X_train, y_train=y_train, problem_type='binary'). When we call .search(), the search for the best pipeline will begin.
AutoMLSearch(X_train=X_train, y_train=y_train, problem_type='binary')
.search()
[5]:
automl = AutoMLSearch(X_train=X_train, y_train=y_train, problem_type='binary', objective=fraud_objective, additional_objectives=['auc', 'f1', 'precision'], max_batches=1, optimize_thresholds=True) automl.search()
Generating pipelines to search over... ***************************** * Beginning pipeline search * ***************************** Optimizing for Fraud Cost. Lower score is better. Searching up to 1 batches for a total of 9 pipelines. Allowed model families: catboost, linear_model, lightgbm, decision_tree, xgboost, extra_trees, random_forest
Batch 1: (1/9) Mode Baseline Binary Classification P... Elapsed:00:00 Starting cross validation Finished cross validation - mean Fraud Cost: 0.033 High coefficient of variation (cv >= 0.2) within cross validation scores. Mode Baseline Binary Classification Pipeline may not perform as estimated on unseen data. Batch 1: (2/9) Logistic Regression Classifier w/ Imp... Elapsed:00:00 Starting cross validation Finished cross validation - mean Fraud Cost: 0.010 High coefficient of variation (cv >= 0.2) within cross validation scores. Logistic Regression Classifier w/ Imputer + One Hot Encoder + Standard Scaler may not perform as estimated on unseen data. Batch 1: (3/9) Random Forest Classifier w/ Imputer +... Elapsed:00:03 Starting cross validation Finished cross validation - mean Fraud Cost: 0.002 Batch 1: (4/9) XGBoost Classifier w/ Imputer + One H... Elapsed:00:05 Starting cross validation Finished cross validation - mean Fraud Cost: 0.010 High coefficient of variation (cv >= 0.2) within cross validation scores. XGBoost Classifier w/ Imputer + One Hot Encoder may not perform as estimated on unseen data. Batch 1: (5/9) CatBoost Classifier w/ Imputer Elapsed:00:06 Starting cross validation Finished cross validation - mean Fraud Cost: 0.002 Batch 1: (6/9) Elastic Net Classifier w/ Imputer + O... Elapsed:00:07 Starting cross validation Finished cross validation - mean Fraud Cost: 0.002 Batch 1: (7/9) Extra Trees Classifier w/ Imputer + O... Elapsed:00:09 Starting cross validation Finished cross validation - mean Fraud Cost: 0.002 Batch 1: (8/9) LightGBM Classifier w/ Imputer + One ... Elapsed:00:10 Starting cross validation Finished cross validation - mean Fraud Cost: 0.033 High coefficient of variation (cv >= 0.2) within cross validation scores. LightGBM Classifier w/ Imputer + One Hot Encoder may not perform as estimated on unseen data. Batch 1: (9/9) Decision Tree Classifier w/ Imputer +... Elapsed:00:12 Starting cross validation Finished cross validation - mean Fraud Cost: 0.012 High coefficient of variation (cv >= 0.2) within cross validation scores. Decision Tree Classifier w/ Imputer + One Hot Encoder may not perform as estimated on unseen data. Search finished after 00:13 Best pipeline: Random Forest Classifier w/ Imputer + One Hot Encoder Best pipeline Fraud Cost: 0.002083
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 call automl.best_pipeline.
automl.best_pipeline
[7]:
best_pipeline = automl.best_pipeline
We can get more details about any pipeline created during the search process, including how it performed on other objective functions, by calling the describe_pipeline method and passing the id of the pipeline of interest.
describe_pipeline
id
[8]:
automl.describe_pipeline(automl.rankings.iloc[1]["id"])
********************************** * CatBoost Classifier w/ Imputer * ********************************** Problem Type: binary Model Family: CatBoost Pipeline Steps ============== 1. Imputer * categorical_impute_strategy : most_frequent * numeric_impute_strategy : mean * categorical_fill_value : None * numeric_fill_value : None 2. CatBoost Classifier * n_estimators : 10 * eta : 0.03 * max_depth : 6 * bootstrap_type : None * silent : True * allow_writing_files : False Training ======== Training for binary problems. Objective to optimize binary classification pipeline thresholds for: <evalml.objectives.fraud_cost.FraudCost object at 0x7efd859cd790> Total training time (including CV): 1.0 seconds Cross Validation ---------------- Fraud Cost AUC F1 Precision # Training # Validation 0 0.002 0.877 0.249 0.142 426.000 267.000 1 0.002 0.817 0.249 0.142 426.000 267.000 2 0.002 0.809 0.244 0.139 427.000 266.000 mean 0.002 0.834 0.248 0.141 - - std 0.000 0.037 0.003 0.002 - - coef of var 0.090 0.045 0.012 0.013 - -
Finally, since the best pipeline is already trained, we evaluate it on the holdout data.
Now, we can score the pipeline on the holdout data using both our fraud cost objective and the AUC (Area under the ROC Curve) objective.
[9]:
best_pipeline.score(X_holdout, y_holdout, objectives=["auc", fraud_objective])
OrderedDict([('AUC', 0.8556893687707641), ('Fraud Cost', 0.0019819080665371956)])
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.
[10]:
automl_auc = AutoMLSearch(X_train=X_train, y_train=y_train, problem_type='binary', objective='auc', additional_objectives=['f1', 'precision'], max_batches=1, optimize_thresholds=True) automl_auc.search()
Generating pipelines to search over... ***************************** * Beginning pipeline search * ***************************** Optimizing for AUC. Greater score is better. Searching up to 1 batches for a total of 9 pipelines. Allowed model families: catboost, linear_model, lightgbm, decision_tree, xgboost, extra_trees, random_forest
Batch 1: (1/9) Mode Baseline Binary Classification P... Elapsed:00:00 Starting cross validation Finished cross validation - mean AUC: 0.500 Batch 1: (2/9) Logistic Regression Classifier w/ Imp... Elapsed:00:00 Starting cross validation Finished cross validation - mean AUC: 0.754 Batch 1: (3/9) Random Forest Classifier w/ Imputer +... Elapsed:00:00 Starting cross validation Finished cross validation - mean AUC: 0.839 Batch 1: (4/9) XGBoost Classifier w/ Imputer + One H... Elapsed:00:01 Starting cross validation Finished cross validation - mean AUC: 0.837 Batch 1: (5/9) CatBoost Classifier w/ Imputer Elapsed:00:02 Starting cross validation Finished cross validation - mean AUC: 0.828 Batch 1: (6/9) Elastic Net Classifier w/ Imputer + O... Elapsed:00:03 Starting cross validation Finished cross validation - mean AUC: 0.500 Batch 1: (7/9) Extra Trees Classifier w/ Imputer + O... Elapsed:00:03 Starting cross validation Finished cross validation - mean AUC: 0.799 Batch 1: (8/9) LightGBM Classifier w/ Imputer + One ... Elapsed:00:04 Starting cross validation Finished cross validation - mean AUC: 0.845 Batch 1: (9/9) Decision Tree Classifier w/ Imputer +... Elapsed:00:05 Starting cross validation Finished cross validation - mean AUC: 0.799 Search finished after 00:06 Best pipeline: LightGBM Classifier w/ Imputer + One Hot Encoder Best pipeline AUC: 0.844609
Like before, we can look at the rankings of all of the pipelines searched and pick the best pipeline.
[11]:
automl_auc.rankings
[12]:
best_pipeline_auc = automl_auc.best_pipeline
[13]:
# get the fraud score on holdout data best_pipeline_auc.score(X_holdout, y_holdout, objectives=["auc", fraud_objective])
OrderedDict([('AUC', 0.9067691029900331), ('Fraud Cost', 0.024416589442515867)])
[14]:
# 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.