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, 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 Targets 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_batches=1, optimize_thresholds=True) automl.search(X_train, y_train)
`X` passed was not a DataTable. EvalML will try to convert the input as a Woodwork DataTable and types will be inferred. To control this behavior, please pass in a Woodwork DataTable instead. `y` passed was not a DataColumn. EvalML will try to convert the input as a Woodwork DataTable and types will be inferred. To control this behavior, please pass in a Woodwork DataTable instead. 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: random_forest, lightgbm, linear_model, extra_trees, catboost, decision_tree, xgboost
Batch 1: (1/9) Mode Baseline Binary Classification P... Elapsed:00:00 Starting cross validation Finished cross validation - mean Fraud Cost: 0.029 Batch 1: (2/9) Decision Tree Classifier w/ Imputer +... Elapsed:00:00 Starting cross validation Finished cross validation - mean Fraud Cost: 0.003 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. Batch 1: (3/9) LightGBM Classifier w/ Imputer + One ... Elapsed:00:02 Starting cross validation Finished cross validation - mean Fraud Cost: 0.016 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: (4/9) Extra Trees Classifier w/ Imputer + O... Elapsed:00:03 Starting cross validation Finished cross validation - mean Fraud Cost: 0.002 Batch 1: (5/9) Elastic Net Classifier w/ Imputer + O... Elapsed:00:06 Starting cross validation Finished cross validation - mean Fraud Cost: 0.002 Batch 1: (6/9) CatBoost Classifier w/ Imputer Elapsed:00:08 Starting cross validation Finished cross validation - mean Fraud Cost: 0.002 Batch 1: (7/9) XGBoost Classifier w/ Imputer + One H... Elapsed:00:09 Starting cross validation Finished cross validation - mean Fraud Cost: 0.002 Batch 1: (8/9) Random Forest Classifier w/ Imputer +... Elapsed:00:10 Starting cross validation Finished cross validation - mean Fraud Cost: 0.002 Batch 1: (9/9) Logistic Regression Classifier w/ Imp... Elapsed:00:13 Starting cross validation Finished cross validation - mean Fraud Cost: 0.002 Search finished after 00:16 Best pipeline: Random Forest Classifier w/ Imputer + One Hot Encoder Best pipeline Fraud Cost: 0.002224
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"])
********************************************************************************* * Logistic Regression Classifier w/ Imputer + One Hot Encoder + Standard Scaler * ********************************************************************************* Problem Type: binary Model Family: Linear Pipeline Steps ============== 1. Imputer * categorical_impute_strategy : most_frequent * numeric_impute_strategy : mean * categorical_fill_value : None * numeric_fill_value : None 2. One Hot Encoder * top_n : 10 * features_to_encode : None * categories : None * drop : None * handle_unknown : ignore * handle_missing : error 3. Standard Scaler 4. Logistic Regression Classifier * penalty : l2 * C : 1.0 * n_jobs : -1 * multi_class : auto * solver : lbfgs Training ======== Training for binary problems. Objective to optimize binary classification pipeline thresholds for: <evalml.objectives.fraud_cost.FraudCost object at 0x7fb473dcb450> Total training time (including CV): 2.7 seconds Cross Validation ---------------- Fraud Cost AUC F1 Precision # Training # Testing 0 0.002 0.804 0.247 0.141 1066.000 667.000 1 0.002 0.821 0.247 0.141 1066.000 667.000 2 0.002 0.797 0.247 0.141 1067.000 666.000 mean 0.002 0.807 0.247 0.141 - - std 0.000 0.012 0.000 0.000 - - coef of var 0.086 0.015 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)
GeneratedPipeline(parameters={'Imputer':{'categorical_impute_strategy': 'most_frequent', 'numeric_impute_strategy': 'mean', 'categorical_fill_value': None, 'numeric_fill_value': None}, 'One Hot Encoder':{'top_n': 10, 'features_to_encode': None, 'categories': None, 'drop': None, 'handle_unknown': 'ignore', 'handle_missing': 'error'}, 'Random Forest Classifier':{'n_estimators': 100, 'max_depth': 6, 'n_jobs': -1},})
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', 0.8370764119601328), ('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_batches=1, optimize_thresholds=True) automl_auc.search(X_train, y_train)
`X` passed was not a DataTable. EvalML will try to convert the input as a Woodwork DataTable and types will be inferred. To control this behavior, please pass in a Woodwork DataTable instead. `y` passed was not a DataColumn. EvalML will try to convert the input as a Woodwork DataTable and types will be inferred. To control this behavior, please pass in a Woodwork DataTable instead. 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: random_forest, lightgbm, linear_model, extra_trees, catboost, decision_tree, xgboost
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) Decision Tree Classifier w/ Imputer +... Elapsed:00:00 Starting cross validation Finished cross validation - mean AUC: 0.816 Batch 1: (3/9) LightGBM Classifier w/ Imputer + One ... Elapsed:00:00 Starting cross validation Finished cross validation - mean AUC: 0.862 Batch 1: (4/9) Extra Trees Classifier w/ Imputer + O... Elapsed:00:01 Starting cross validation Finished cross validation - mean AUC: 0.812 Batch 1: (5/9) Elastic Net Classifier w/ Imputer + O... Elapsed:00:03 Starting cross validation Finished cross validation - mean AUC: 0.500 Batch 1: (6/9) CatBoost Classifier w/ Imputer Elapsed:00:04 Starting cross validation Finished cross validation - mean AUC: 0.826 Batch 1: (7/9) XGBoost Classifier w/ Imputer + One H... Elapsed:00:04 Starting cross validation Finished cross validation - mean AUC: 0.857 Batch 1: (8/9) Random Forest Classifier w/ Imputer +... Elapsed:00:05 Starting cross validation Finished cross validation - mean AUC: 0.855 Batch 1: (9/9) Logistic Regression Classifier w/ Imp... Elapsed:00:07 Starting cross validation Finished cross validation - mean AUC: 0.805 Search finished after 00:08 Best pipeline: LightGBM Classifier w/ Imputer + One Hot Encoder Best pipeline AUC: 0.861867
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)
GeneratedPipeline(parameters={'Imputer':{'categorical_impute_strategy': 'most_frequent', 'numeric_impute_strategy': 'mean', 'categorical_fill_value': None, 'numeric_fill_value': None}, 'One Hot Encoder':{'top_n': 10, 'features_to_encode': None, 'categories': None, 'drop': None, 'handle_unknown': 'ignore', 'handle_missing': 'error'}, 'LightGBM Classifier':{'boosting_type': 'gbdt', 'learning_rate': 0.1, 'n_estimators': 100, 'max_depth': 0, 'num_leaves': 31, 'min_child_samples': 20, 'n_jobs': -1, 'bagging_freq': 0, 'bagging_fraction': 0.9},})
[14]:
# get the fraud score on holdout data best_pipeline_auc.score(X_holdout, y_holdout, objectives=["auc", fraud_objective])
OrderedDict([('AUC', 0.8441860465116278), ('Fraud Cost', 0.004348876090615287)])
[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.