Building a Lead Scoring Model with EvalML¶
In this demo, we will build an optimized lead scoring model using EvalML. To optimize the pipeline, we will set up an objective function to maximize the revenue generated with true positives while taking into account the cost of false positives. At the end of this demo, we also show you how introducing the right objective during the training is significantly better than using a generic machine learning metric like AUC.
[1]:
import evalml
from evalml import AutoMLSearch
from evalml.objectives import LeadScoring
Configure LeadScoring¶
To optimize the pipelines toward the specific business needs of this model, you can set your own assumptions for how much value is gained through true positives and the cost associated with false positives. These parameters are
true_positive- dollar amount to be gained with a successful leadfalse_positive- dollar amount to be lost with an unsuccessful lead
Using these parameters, EvalML builds a pileline that will maximize the amount of revenue per lead generated.
[2]:
lead_scoring_objective = LeadScoring(
true_positives=1000,
false_positives=-10
)
Dataset¶
We will be utilizing a dataset detailing a customer’s job, country, state, zip, online action, the dollar amount of that action and whether they were a successful lead.
[3]:
from urllib.request import urlopen
import pandas as pd
import woodwork as ww
customers_data = urlopen('https://featurelabs-static.s3.amazonaws.com/lead_scoring_ml_apps/customers.csv')
interactions_data = urlopen('https://featurelabs-static.s3.amazonaws.com/lead_scoring_ml_apps/interactions.csv')
leads_data = urlopen('https://featurelabs-static.s3.amazonaws.com/lead_scoring_ml_apps/previous_leads.csv')
customers = pd.read_csv(customers_data)
interactions = pd.read_csv(interactions_data)
leads = pd.read_csv(leads_data)
X = customers.merge(interactions, on='customer_id').merge(leads, on='customer_id')
y = X['label']
X = X.drop(['customer_id', 'date_registered', 'birthday','phone', 'email',
'owner', 'company', 'id', 'time_x',
'session', 'referrer', 'time_y', 'label', 'country'], axis=1)
display(X.head())
| job | state | zip | action | amount | |
|---|---|---|---|---|---|
| 0 | Engineer, mining | NY | 60091.0 | page_view | NaN |
| 1 | Psychologist, forensic | CA | NaN | purchase | 135.23 |
| 2 | Psychologist, forensic | CA | NaN | page_view | NaN |
| 3 | Air cabin crew | NaN | 60091.0 | download | NaN |
| 4 | Air cabin crew | NaN | 60091.0 | page_view | NaN |
We will convert our data into Woodwork data structures. Doing so enables us to have more control over the types passed to and inferred by AutoML.
[4]:
X.ww.init(semantic_tags={'job': 'category'}, logical_types={'job': 'Categorical'})
y = ww.init_series(y)
X.ww
[4]:
| Physical Type | Logical Type | Semantic Tag(s) | |
|---|---|---|---|
| Column | |||
| job | category | Categorical | ['category'] |
| state | category | Categorical | ['category'] |
| zip | float64 | Double | ['numeric'] |
| action | category | Categorical | ['category'] |
| amount | float64 | Double | ['numeric'] |
Search for the best pipeline¶
In order to validate the results of the pipeline creation and optimization process, we will save some of our data as a holdout set.
EvalML natively supports one-hot encoding and imputation so the above NaN and categorical values will be taken care of.
[5]:
X_train, X_holdout, y_train, y_holdout = evalml.preprocessing.split_data(X, y, problem_type='binary', test_size=0.2, random_seed=0)
X.ww
[5]:
| Physical Type | Logical Type | Semantic Tag(s) | |
|---|---|---|---|
| Column | |||
| job | category | Categorical | ['category'] |
| state | category | Categorical | ['category'] |
| zip | float64 | Double | ['numeric'] |
| action | category | Categorical | ['category'] |
| amount | float64 | Double | ['numeric'] |
Because the lead scoring 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.
[6]:
automl = AutoMLSearch(X_train=X_train, y_train=y_train,
problem_type='binary',
objective=lead_scoring_objective,
additional_objectives=['auc'],
max_batches=1,
optimize_thresholds=True,
sampler_method=None)
automl.search()
Generating pipelines to search over...
8 pipelines ready for search.
*****************************
* Beginning pipeline search *
*****************************
Optimizing for Lead Scoring.
Greater score is better.
Using SequentialEngine to train and score pipelines.
Searching up to 1 batches for a total of 9 pipelines.
Allowed model families: extra_trees, decision_tree, lightgbm, linear_model, xgboost, catboost, random_forest
Evaluating Baseline Pipeline: Mode Baseline Binary Classification Pipeline
Mode Baseline Binary Classification Pipeline:
Starting cross validation
Finished cross validation - mean Lead Scoring: 0.000
*****************************
* Evaluating Batch Number 1 *
*****************************
Elastic Net Classifier w/ Imputer + One Hot Encoder + Standard Scaler:
Starting cross validation
Finished cross validation - mean Lead Scoring: 15.550
Decision Tree Classifier w/ Imputer + One Hot Encoder:
Starting cross validation
Finished cross validation - mean Lead Scoring: 14.821
Random Forest Classifier w/ Imputer + One Hot Encoder:
Starting cross validation
Finished cross validation - mean Lead Scoring: 14.515
LightGBM Classifier w/ Imputer + One Hot Encoder:
Starting cross validation
Finished cross validation - mean Lead Scoring: 13.354
Logistic Regression Classifier w/ Imputer + One Hot Encoder + Standard Scaler:
Starting cross validation
Finished cross validation - mean Lead Scoring: 14.941
XGBoost Classifier w/ Imputer + One Hot Encoder:
Starting cross validation
Finished cross validation - mean Lead Scoring: 14.537
Extra Trees Classifier w/ Imputer + One Hot Encoder:
Starting cross validation
Finished cross validation - mean Lead Scoring: 15.294
CatBoost Classifier w/ Imputer:
Starting cross validation
Finished cross validation - mean Lead Scoring: 15.997
Search finished after 00:17
Best pipeline: CatBoost Classifier w/ Imputer
Best pipeline Lead Scoring: 15.996914
View rankings and select pipeline¶
Once the fitting process is done, we can see all of the pipelines that were searched, ranked by their score on the lead scoring objective we defined.
[7]:
automl.rankings
[7]:
| id | pipeline_name | mean_cv_score | standard_deviation_cv_score | validation_score | percent_better_than_baseline | high_variance_cv | parameters | |
|---|---|---|---|---|---|---|---|---|
| 0 | 8 | CatBoost Classifier w/ Imputer | 15.996914 | 0.645669 | 16.741935 | inf | False | {'Imputer': {'categorical_impute_strategy': 'm... |
| 1 | 1 | Elastic Net Classifier w/ Imputer + One Hot En... | 15.549506 | 0.300398 | 15.741935 | inf | False | {'Imputer': {'categorical_impute_strategy': 'm... |
| 2 | 7 | Extra Trees Classifier w/ Imputer + One Hot En... | 15.293603 | 0.436765 | 14.877419 | inf | False | {'Imputer': {'categorical_impute_strategy': 'm... |
| 3 | 5 | Logistic Regression Classifier w/ Imputer + On... | 14.940900 | 0.914874 | 13.929032 | inf | False | {'Imputer': {'categorical_impute_strategy': 'm... |
| 4 | 2 | Decision Tree Classifier w/ Imputer + One Hot ... | 14.820570 | 2.443147 | 16.741935 | inf | False | {'Imputer': {'categorical_impute_strategy': 'm... |
| 5 | 6 | XGBoost Classifier w/ Imputer + One Hot Encoder | 14.536504 | 1.124271 | 15.541935 | inf | False | {'Imputer': {'categorical_impute_strategy': 'm... |
| 6 | 3 | Random Forest Classifier w/ Imputer + One Hot ... | 14.514975 | 0.504420 | 14.948387 | inf | False | {'Imputer': {'categorical_impute_strategy': 'm... |
| 7 | 4 | LightGBM Classifier w/ Imputer + One Hot Encoder | 13.353574 | 1.575213 | 14.400000 | inf | False | {'Imputer': {'categorical_impute_strategy': 'm... |
| 8 | 0 | Mode Baseline Binary Classification Pipeline | 0.000000 | 0.000000 | 0.000000 | 0.0 | False | {'Baseline Classifier': {'strategy': 'mode'}} |
To select the best pipeline we can call automl.best_pipeline.
[8]:
best_pipeline = automl.best_pipeline
Describe pipeline¶
You can get more details about any pipeline, including how it performed on other objective functions by calling .describe_pipeline() and specifying the id of the pipeline.
[9]:
automl.describe_pipeline(automl.rankings.iloc[0]["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.lead_scoring.LeadScoring object at 0x7f1d28cdd2b0>
Total training time (including CV): 0.9 seconds
Cross Validation
----------------
Lead Scoring AUC # Training # Validation
0 16.742 0.600 3,099 1,550
1 15.600 0.509 3,099 1,550
2 15.649 0.576 3,100 1,549
mean 15.997 0.562 - -
std 0.646 0.047 - -
coef of var 0.040 0.084 - -
Evaluate on hold out¶
Finally, since the best pipeline was trained on all of the training data, we evaluate it on the holdout dataset.
[10]:
best_pipeline.score(X_holdout, y_holdout, objectives=["auc", lead_scoring_objective])
[10]:
OrderedDict([('AUC', 0.559035962526443), ('Lead Scoring', 15.382631126397248)])
Why optimize for a problem-specific objective?¶
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 lead scoring objective to see how the best pipelines compare.
[11]:
automl_auc = evalml.AutoMLSearch(X_train=X_train, y_train=y_train,
problem_type='binary',
objective='auc',
additional_objectives=[],
max_batches=1,
optimize_thresholds=True,
sampler_method=None)
automl_auc.search()
Generating pipelines to search over...
8 pipelines ready for search.
*****************************
* Beginning pipeline search *
*****************************
Optimizing for AUC.
Greater score is better.
Using SequentialEngine to train and score pipelines.
Searching up to 1 batches for a total of 9 pipelines.
Allowed model families: extra_trees, decision_tree, lightgbm, linear_model, xgboost, catboost, random_forest
Evaluating Baseline Pipeline: Mode Baseline Binary Classification Pipeline
Mode Baseline Binary Classification Pipeline:
Starting cross validation
Finished cross validation - mean AUC: 0.500
*****************************
* Evaluating Batch Number 1 *
*****************************
Elastic Net Classifier w/ Imputer + One Hot Encoder + Standard Scaler:
Starting cross validation
Finished cross validation - mean AUC: 0.630
Decision Tree Classifier w/ Imputer + One Hot Encoder:
Starting cross validation
Finished cross validation - mean AUC: 0.594
Random Forest Classifier w/ Imputer + One Hot Encoder:
Starting cross validation
Finished cross validation - mean AUC: 0.682
LightGBM Classifier w/ Imputer + One Hot Encoder:
Starting cross validation
Finished cross validation - mean AUC: 0.671
Logistic Regression Classifier w/ Imputer + One Hot Encoder + Standard Scaler:
Starting cross validation
Finished cross validation - mean AUC: 0.681
XGBoost Classifier w/ Imputer + One Hot Encoder:
Starting cross validation
Finished cross validation - mean AUC: 0.701
Extra Trees Classifier w/ Imputer + One Hot Encoder:
Starting cross validation
Finished cross validation - mean AUC: 0.689
CatBoost Classifier w/ Imputer:
Starting cross validation
Finished cross validation - mean AUC: 0.542
Search finished after 00:07
Best pipeline: XGBoost Classifier w/ Imputer + One Hot Encoder
Best pipeline AUC: 0.701032
[12]:
automl_auc.rankings
[12]:
| id | pipeline_name | mean_cv_score | standard_deviation_cv_score | validation_score | percent_better_than_baseline | high_variance_cv | parameters | |
|---|---|---|---|---|---|---|---|---|
| 0 | 6 | XGBoost Classifier w/ Imputer + One Hot Encoder | 0.701032 | 0.024980 | 0.720642 | 20.103249 | False | {'Imputer': {'categorical_impute_strategy': 'm... |
| 1 | 7 | Extra Trees Classifier w/ Imputer + One Hot En... | 0.689375 | 0.036617 | 0.727598 | 18.937464 | False | {'Imputer': {'categorical_impute_strategy': 'm... |
| 2 | 3 | Random Forest Classifier w/ Imputer + One Hot ... | 0.682025 | 0.050716 | 0.732789 | 18.202538 | False | {'Imputer': {'categorical_impute_strategy': 'm... |
| 3 | 5 | Logistic Regression Classifier w/ Imputer + On... | 0.680759 | 0.013717 | 0.696594 | 18.075908 | False | {'Imputer': {'categorical_impute_strategy': 'm... |
| 4 | 4 | LightGBM Classifier w/ Imputer + One Hot Encoder | 0.671060 | 0.028763 | 0.698287 | 17.105985 | False | {'Imputer': {'categorical_impute_strategy': 'm... |
| 5 | 1 | Elastic Net Classifier w/ Imputer + One Hot En... | 0.629855 | 0.003297 | 0.626837 | 12.985451 | False | {'Imputer': {'categorical_impute_strategy': 'm... |
| 6 | 2 | Decision Tree Classifier w/ Imputer + One Hot ... | 0.594215 | 0.006959 | 0.601705 | 9.421468 | False | {'Imputer': {'categorical_impute_strategy': 'm... |
| 7 | 8 | CatBoost Classifier w/ Imputer | 0.542394 | 0.039310 | 0.505055 | 4.239420 | False | {'Imputer': {'categorical_impute_strategy': 'm... |
| 8 | 0 | Mode Baseline Binary Classification Pipeline | 0.500000 | 0.000000 | 0.500000 | 0.000000 | False | {'Baseline Classifier': {'strategy': 'mode'}} |
Like before, we can look at the rankings and pick the best pipeline.
[13]:
best_pipeline_auc = automl_auc.best_pipeline
[14]:
# get the auc and lead scoring score on holdout data
best_pipeline_auc.score(X_holdout, y_holdout, objectives=["auc", lead_scoring_objective])
[14]:
OrderedDict([('AUC', 0.6662964641885766),
('Lead Scoring', -0.008598452278589854)])
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 lead scoring. However, the revenue per lead is much smaller per lead when optimized for AUC and was much larger when optimized for lead scoring. As a result, we would have a huge gain on the amount of revenue if we optimized for lead scoring.
This happens because optimizing for AUC does not take into account the user-specified true_positive (dollar amount to be gained with a successful lead) and false_positive (dollar amount to be lost with an unsuccessful lead) values. Thus, the best pipelines may produce the highest AUC but may not actually generate the most revenue through lead scoring.
This example highlights how performance in the real world can diverge greatly from machine learning metrics.