Driver-s-Race-Classifer-traffic-stops

To see if drivers were being profiled. I built a Support Vector Machine (SVM) classifier and a randomForest classifier to predict a driver's race given the traffic's stop's details. Successful classification will indicate the existence of bais in the traffic stops' data.

Data

Two Data Tables:

1. Traffic Stop Data
2. Socio Economic Data for Zip Codes where Stops happen

Traffic Stop Data

Unnamed: 0	Stop_Key	Type	TCOLE_Sex	TCOLE_RACE_ETHNICITY	Standardized_Race_Known	Reason_for_Stop	Street_Type	Search_Yes_or_No	TCOLE_Search_Based_On	TCOLE_Search_Found	TCOLE_Result_of_Stop	TCOLE_Arrest_Based_On	Council_District	COUNTY	Custody	Location	Sector	Standardized_Race	Stop_Time
0	20201-459626502-25962	WARNING	Male	White	NO - RACE OR ETHNICITY WAS NOT KNOWN BEFORE STOP	Moving Traffic Violation	City Street	NO	N/A - No Search was conducted	N/A - No Search was conducted	Verbal Warning	N/A - No Arrest Conducted	Council District 9	TRAVIS COUNTY	NOT APPLICABLE	500 E 8TH ST	GEORGE	WHITE	2154.0

Socio Economic Data for Zip Codes

index	Zip_Code_1	Zip_Code	latitude	longitude	propertyTaxRate	numPriceChanges	avgSchoolRating	MedianStudentsPerTeacher
0	78617	78617	30.16451458598292	-97.63406638211984	1.9799999999999984	2.558139534883721	3.1589147286821717	13.965116279069768
1	78619	78619	30.136290550231934	-97.97578048706056	2.01	1.9166666666666667	7.388888888888889	15.666666666666666

Joined (merged) the Tables

racialProf = racialProfUpdated.merge(socioEcoZipCodesInfo, on ='Zip_Code', how = 'outer')

index	Unnamed: 0	Stop_Key	Type	TCOLE_Sex	TCOLE_RACE_ETHNICITY	Standardized_Race_Known	Reason_for_Stop	Street_Type	Search_Yes_or_No	TCOLE_Search_Based_On	TCOLE_Search_Found	TCOLE_Result_of_Stop	TCOLE_Arrest_Based_On	Council_District	COUNTY	Custody	Location	Sector	Standardized_Race	Stop_Time
0	0.0	20201-459626502-25962	WARNING	Male	White	NO - RACE OR ETHNICITY WAS NOT KNOWN BEFORE STOP	Moving Traffic Violation	City Street	NO	N/A - No Search was conducted	N/A - No Search was conducted	Verbal Warning	N/A - No Arrest Conducted	Council District 9	TRAVIS COUNTY	NOT APPLICABLE	500 E 8TH ST	GEORGE	WHITE	2154.0

Shape: (45274, 33)

Data Cleaning Pipeline: Encoder and Imputer

numeric_transformer = Pipeline(steps=[('imputer', SimpleImputer(strategy='median')),('scaler', StandardScaler())])
categorical_transformer = Pipeline(steps=[('imputer', SimpleImputer(strategy='constant',fill_value='missing')),
                                          ('onehot', OneHotEncoder(handle_unknown='ignore'))])
numeric_features = train.select_dtypes(include=['int64','float64']).columns
categorical_features = train.select_dtypes(include=['object']).drop(['TCOLE_RACE_ETHNICITY'], axis=1).columns
preprocessor = ColumnTransformer( transformers=[('num', numeric_transformer, numeric_features),
                                                ('cat', categorical_transformer, categorical_features)])

Train Test Split

train, test, = train_test_split(racialProf,
test_size=0.2)

RandomForest Model

rf = Pipeline(steps=[('preprocessor', preprocessor),
('classifier', RandomForestClassifier(n_estimators = 13, max_depth=10))])

Training Fit

X_train = train.drop('TCOLE_RACE_ETHNICITY', axis=1)
y_train = train['TCOLE_RACE_ETHNICITY']
rf.fit(X_train ,y_train)
train_pred = rf.predict (X_train)

Test Fit

X_test = test.drop('TCOLE_RACE_ETHNICITY', axis=1)
y_test = test['TCOLE_RACE_ETHNICITY']
test_predRF = rf.predict (X_test)

RF Grid Search

param_grid = {
'classifier__n_estimators': [200, 500],
'classifier__max_features': ['auto', 'sqrt', 'log2'],
'classifier__max_depth' : [4,5,6,7,8],
'classifier__criterion' :['gini', 'entropy']}
from sklearn.model_selection import GridSearchCV
CV = GridSearchCV(rf, param_grid, n_jobs= 1)
CV.fit(X_train, y_train)
print(CV.best_params_)
print(CV.best_score_)

# Suggested parameters from grid search
from sklearn.ensemble import RandomForestClassifier
rf_GridSearch = Pipeline(steps=[('preprocessor', preprocessor),
('classifier', RandomForestClassifier(criterion = 'entropy', 
                                      max_depth = 8, max_features= 'sqrt',
                                        n_estimators= 200))])

RF Grid Search Training Fit

rf_GridSearch.fit(X_train, y_train)

RF Grid Search Test Fit

test_predRFGRID = rf_GridSearch.predict (X_test)

SVM Model

svm_classifier = Pipeline(steps=[('preprocessor', preprocessor),
('classifier', SVC(kernel="poly", degree=3))])

Training Fit

X_train = train.drop('TCOLE_RACE_ETHNICITY', axis=1)
y_train = train['TCOLE_RACE_ETHNICITY']
svm_classifier.fit(X_train, y_train)
train_predSVM = svm_classifier.predict (X_train)

Test Fit

X_test = test.drop('TCOLE_RACE_ETHNICITY', axis=1)
y_test = test['TCOLE_RACE_ETHNICITY']
test_predSVM = svm_classifier.predict (X_test)

SVM Grid Search

param_grid = {
'classifier__kernel': ['poly'],
'classifier__degree': [1,2,3],
'classifier__decision_function_shape': ['ovr','ovo']}
from sklearn.model_selection import GridSearchCV
CV1 = GridSearchCV(svm_classifier, param_grid, n_jobs= 1)
CV1.fit(X_train, y_train)
print(CV1.best_params_)
print(CV1.best_score_)

# Suggested Parameters from Grid-Search
svm_classifierGridSearch = Pipeline(steps=[('preprocessor', preprocessor),
('classifier', SVC(decision_function_shape= 'ovr', degree= 1, kernel= 'poly'
))])

SVM Grid Search Training Fit

svm_classifierGridSearch.fit(X_train, y_train)
train_predSVMGRID = svm_classifierGridSearch.predict (X_train)

SVM Grid Search Test Fit

test_predSVMGRID = svm_classifierGridSearch.predict (X_test)