Skip to content

Insurance companies invest a lot of time and money into optimizing their pricing and accurately estimating the likelihood that customers will make a claim. In many countries insurance it is a legal requirement to have car insurance in order to drive a vehicle on public roads, so the market is very large!

(Source: https://www.accenture.com/_acnmedia/pdf-84/accenture-machine-leaning-insurance.pdf)

Knowing all of this, On the Road car insurance have requested your services in building a model to predict whether a customer will make a claim on their insurance during the policy period. As they have very little expertise and infrastructure for deploying and monitoring machine learning models, they've asked you to identify the single feature that results in the best performing model, as measured by accuracy, so they can start with a simple model in production.

They have supplied you with their customer data as a csv file called car_insurance.csv, along with a table detailing the column names and descriptions below.

The dataset

ColumnDescription
idUnique client identifier
ageClient's age:
  • 0: 16-25
  • 1: 26-39
  • 2: 40-64
  • 3: 65+
genderClient's gender:
  • 0: Female
  • 1: Male
driving_experienceYears the client has been driving:
  • 0: 0-9
  • 1: 10-19
  • 2: 20-29
  • 3: 30+
educationClient's level of education:
  • 0: No education
  • 1: High school
  • 2: University
incomeClient's income level:
  • 0: Poverty
  • 1: Working class
  • 2: Middle class
  • 3: Upper class
credit_scoreClient's credit score (between zero and one)
vehicle_ownershipClient's vehicle ownership status:
  • 0: Does not own their vehilce (paying off finance)
  • 1: Owns their vehicle
vehcile_yearYear of vehicle registration:
  • 0: Before 2015
  • 1: 2015 or later
marriedClient's marital status:
  • 0: Not married
  • 1: Married
childrenClient's number of children
postal_codeClient's postal code
annual_mileageNumber of miles driven by the client each year
vehicle_typeType of car:
  • 0: Sedan
  • 1: Sports car
speeding_violationsTotal number of speeding violations received by the client
duisNumber of times the client has been caught driving under the influence of alcohol
past_accidentsTotal number of previous accidents the client has been involved in
outcomeWhether the client made a claim on their car insurance (response variable):
  • 0: No claim
  • 1: Made a claim
# Import required modules
import pandas as pd
import numpy as np
from statsmodels.formula.api import logit

car_insurance = pd.read_csv('car_insurance.csv')
car_insurance.head()
# Start coding!

The columns "credit_score" and "annual_mileage" have missing values based on the count of values comparing it to the count of values of "id" column

car_insurance.describe()
cols_missing = ['credit_score','annual_mileage']

for col in cols_missing:
    skewness = car_insurance[col].skew()
    if skewness <= 0.25 and skewness >= -0.25: # condition for mean imputation, describes near symmetrical distribution
        print(f'{col} skewness: {skewness}\nMean imputation performed')
        car_insurance[col] = car_insurance[col].fillna(car_insurance[col].mean())
    else: # condition for median imputation, describes moderate to highly skewed distribution
        print(f'{col} skewness: {skewness}\nMedian imputation performed')
        car_insurance[col] = car_insurance[col].fillna(car_insurance[col].median())
car_insurance.describe()
car_insurance.info()

Check for inconsistencies in values

for col in car_insurance.columns:
    unique_values = car_insurance[col].unique()
    if len(unique_values) <= 50:
        print(f'{col} --> {unique_values}')

Remapping data types, values, and categorical ordering

replace_map = {
    'driving_experience': {'0-9y':0, '10-19y':1, '20-29y':2, '30y+':3},
    'education': {'high school':1, 'none':0, 'university':2},
    'income': {'upper class':3, 'poverty':0, 'working class':1, 'middle class':2},
    'vehicle_year': {'before 2015':0, 'after 2015':1},
    'vehicle_type': {'sedan':0, 'sports car':1}
}

dtype_map = {
    'category':['age', 'driving_experience', 'education', 'income'],
    'bool':['gender', 'vehicle_ownership', 'vehicle_year', 'married', 'vehicle_type'],
    'int32':['children','postal_code','speeding_violations','duis','past_accidents', 'outcome'],
    'float16':['credit_score','annual_mileage']
}

orderedcategory_map = {
    'age': [0, 1, 2, 3],
    'driving_experience': [0, 1, 2, 3],
    'education': [0, 1, 2],
    'income': [0, 1, 2, 3]
}

for col, values_map in replace_map.items():
    car_insurance[col] = car_insurance[col].replace(values_map)

for dtype, cols in dtype_map.items():
    for col in cols:
        car_insurance[col] = car_insurance[col].astype(dtype)

for col, order in orderedcategory_map.items():
    dtype = pd.CategoricalDtype(categories=order, ordered=True)
    car_insurance[col] = car_insurance[col].astype(dtype)

car_insurance.info()
for col in car_insurance.columns:
    unique_values = car_insurance[col].unique()
    if len(unique_values) <= 50:
        print(f'{col} --> {unique_values}')
features = car_insurance.columns.drop(['id','outcome'])
features
models_list = []

for feature in features:
    model = logit(f'outcome ~ {feature}', data = car_insurance).fit()
    CfMtx = model.pred_table()
    accuracy = (CfMtx[0][0] + CfMtx[1][1]) / CfMtx.sum()
    models_list.append([feature, accuracy])
models_list_sorted = sorted(models_list, key=lambda x: x[1], reverse=True)
models_list_sorted

best_feature_df = pd.DataFrame({'best_feature':[models_list_sorted[0][0]],
                                'best_accuracy':[models_list_sorted[0][1]]})

best_feature_df