Introduction
The purpose of this project is to leverage machine learning techniques to predict the probability of customers purchasing vehicle insurance based on historical data provided by a large International Insurance company. The dataset contains information about the attempts made by the company to cross-sell car insurance to its current medical insurance policyholders. The objective is to build models that can accurately classify customers into two categories: those who purchased the insurance and those who did not.
Objective
The primary goal of this project is to predict the probability of a customer purchasing car insurance within 30 days post-contact. This prediction is vital for the sales representatives of the company to optimize their efforts and tailor their strategies towards potential customers who are more likely to make a purchase. The project aims to apply various machine learning models to achieve the highest possible predictive accuracy, measured by the AUC ROC (Area Under the Curve) and F-1 score.
About the Dataset
The dataset provided by the insurance company includes several features such as customer demographics (age, gender), policy information (tenure, region), driving-related details (driver's license status, existing car insurance), vehicle characteristics (age, accident history, premium quote), and details about the sales representative involved in the interaction. The target variable, "buy," indicates whether the customer purchased the car insurance (1) or not (0).
Project Flow
The project will follow a structured flow, beginning with the data preparation phase. This involves importing the dataset, performing a train-test split, and evaluating the validity of the split. The subsequent step is the Exploratory Data Analysis (EDA), where descriptive summaries, data manipulations, and visualizations will be conducted to gain insights into the dataset. While analyzing missing values is not mandatory, it is essential to explore potential outliers and erroneous values.
The third phase of the project is dedicated to the development and assessment of the machine learning model. We will train a variety of models on our dataset, specifically utilizing Random Forest Classifier, Logistic Regression, Support Vector Classifier, and Gradient Boosting Classifier. Our evaluation will focus on metrics such as the AUC-ROC curve and the F-1 score to determine the most effective model. The final chosen model will be applied to the Score dataset to estimate the likelihood of customers purchasing car insurance. The predictive outcomes will subsequently be compiled into a CSV file.