Predictive Medical Costs: A Data Analysis and Machine Learning Approach

This case study explores a data-driven approach to predicting medical costs using demographic, lifestyle, and derived factors such as BMI category, age group, smoker status, family size, risk group, and cost ranges. Using the Medical Cost Dataset from Kaggle, the project involved data cleaning, feature engineering, exploratory analysis, and building machine learning models—specifically Linear Regression and Random Forest—to forecast medical expenses. The analysis highlights key cost drivers such as smoking, BMI, and risk group, emphasizing preventive healthcare's role in reducing expenses. This work demonstrates how predictive modeling can provide actionable insights to address healthcare cost challenges effectively.

---

Table of Contents

1. Ask: Problem Statement
2. Prepare: Dataset & Understanding the Data
3. Process: Data Cleaning & Preprocessing
4. Analyze: Exploratory Data Analysis (EDA)
5. Model: Predicting Medical Charges with Regression
6. Share: Interactive Dashboard & Visualization
7. Act: Insights and Future Improvements
8. Project Resources

---

1. Ask: Problem Statement

Healthcare costs are a significant concern for individuals and insurance companies. This project aims to predict medical costs based on demographic, lifestyle, and derived factors such as age, BMI, smoking status, and new categorical features. By leveraging machine learning techniques, the goal is to develop a model that predicts medical expenses for individuals.

• Target Variable: charges (Medical cost in USD)
• Objective: Build a regression model to predict medical charges based on individual features and engineered categories.

---

2. Prepare: Dataset & Understanding the Data

The dataset used for this project is the Medical Cost Dataset on Kaggle, which includes demographic and medical information about individuals.

Features:

• Age: Age of the individual
• Sex: Gender of the individual
• BMI: Body Mass Index
• Children: Number of children/dependents
• Smoker_binary: Encoded smoker status into 0 (Non-smoker) and 1 (Smoker).
• Region: Geographical region (e.g., Southeast, Southwest, etc.)

Derived Features:

1. BMI_category: Categorized BMI into Underweight, Normal, Overweight, and Obese.
2. Age_category: Grouped ages into categories such as Young, Middle-aged, and Senior.
3. Smoker: Whether the individual is a smoker (Yes/No)
4. Family_size: Derived from the children column into categories like No children and Large family.
5. Risk_group: Categorized individuals into Low-risk, Medium-risk, and High-risk based on smoking and BMI.
6. Cost_ranges: Grouped medical charges into ranges such as Low, Medium, and High.

Target Variable:

• charges: The medical expenses in USD that an individual incurs.

---

3. Process: Data Cleaning & Preprocessing

To prepare the data for analysis, several preprocessing steps were performed:

1. Data Cleaning: Removed irrelevant or duplicate columns and handled missing values.
2. Feature Engineering:
   • Derived new features like BMI_category, Age_category, and Risk_group.
   • Converted categorical features (sex, smoker, region) and derived features into numerical values.
   • Checked correlations between features and medical charges.
3. Feature Scaling: Standardized numerical features such as age and BMI to improve model performance.
4. Data Splitting: Divided the dataset into training (80%) and testing (20%) sets for model evaluation.

---

4. Analyze: Exploratory Data Analysis (EDA)

Exploratory analysis was performed to uncover insights and relationships between original and derived features.

Objectives:

• Understand data distribution and relationships.
• Identify significant predictors of medical costs.
• Visualize the data using histograms, scatterplots, bar charts, and correlation heatmaps.

Key Findings from EDA:

1. Smokers incur significantly higher medical costs than non-smokers.
2. Individuals categorized as Obese have higher average costs than those in lower BMI categories.
3. High-risk groups, as defined by the risk group feature, represent the highest average medical charges.

Key Libraries: ggplot2, dplyr, tidyr.

5. Model: Predicting Medical Charges with Regression

Two models were trained to predict medical costs based on the identified features:

Models Trained:

1. Linear Regression:

   • RMSE: 5521.50
   • MAE: 3742.80
   • R-squared: 0.761

2. Random Forest:

   • RMSE: 5514.34
   • MAE: 3944.00
   • R-squared: 0.754

Key Takeaways:

• Both models performed comparably, with Linear Regression slightly outperforming Random Forest in R-squared.
• Smoker_binary, BMI_category, and Risk_group were identified as the most significant predictors of higher medical costs.
  

---

6. Share: Interactive Dashboard & Visualization

An interactive Tableau Dashboard is under development to showcase:

• Predictions and model evaluation metrics.
• Key feature importance, including derived categories like Risk_group and BMI_category.
• Interactive exploration of medical cost trends based on user-selected features.
  Interactive Tableau Dashboard

---

7. Act: Insights and Future Improvements

Key Insights:

1. Smoking is a major contributor to higher medical costs, indicating a strong need for targeted smoking cessation programs.
2. High-risk individuals consistently face higher medical charges, underscoring the importance of preventive care and personalized interventions.
3. Categorical features like BMI category and Risk group offer valuable insights into medical cost patterns, making the model more interpretable and actionable. Future Directions:

Feature Expansion:

1. Include additional features such as family history, lifestyle factors, and socioeconomic status to enhance model accuracy.
2. Explore advanced feature engineering (e.g., interaction terms, transformations) to uncover more complex data relationships.

Model Improvement:

1. Experiment with advanced algorithms like XGBoost and Neural Networks for better prediction performance.
2. Fine-tune models with hyperparameter optimization to improve accuracy.

Dashboard Enhancement:

1. Improve interactive visualizations for better user experience and more actionable insights.

---

8. Project Resources

• Dataset: Medical Cost Dataset on Kaggle
• Cleaned Data: Google Sheets Link
• Tableau Dashboard: Interactive Tableau Dashboard