Medical Cost Analysis: Prediction & Classification

This project leverages supervised learning techniques to analyze a medical cost dataset. It aims to both predict continuous medical charges and classify costs into distinct categories.

Project Overview

The core of this project involves:

• Data Loading and Exploration: Initial steps include loading the dataset and performing exploratory data analysis (EDA) with various visualizations to understand the data's characteristics.
• Model Building: Developing and evaluating machine learning models for two distinct tasks: regression and classification.

Models Used

1. Linear Regression (for predicting 'charges')

Purpose: To predict continuous medical charges based on patient attributes like age, BMI, smoker status, and more.

Implementation Details: A sklearn.pipeline.Pipeline is constructed to streamline the data processing and model training. This pipeline includes:

• Preprocessing:
  • sklearn.preprocessing.OneHotEncoder for nominal categorical features (sex, smoker, region).
  • sklearn.preprocessing.StandardScaler for numerical features (age, BMI, children) to standardize their scale.
  • sklearn.impute.SimpleImputer to handle any missing values in the dataset.
• Model: A sklearn.linear_model.LinearRegression model trained on the preprocessed data.

Evaluation: The model's performance is assessed using Mean Squared Error (MSE) and R-squared (R²) on both the training and test datasets.

2. Classification Models (for predicting 'charges_group')

Purpose: To categorize medical costs into three distinct groups (low, medium, and high) based on the same patient features.

Implementation Details:

• A new categorical feature, charges_group, is created by binning the continuous 'charges' into three defined cost categories.
• Similar preprocessing steps to the Linear Regression pipeline are applied to the data.
• The following classification algorithms are trained and evaluated:
  • sklearn.tree.DecisionTreeClassifier
  • sklearn.ensemble.RandomForestClassifier
  • sklearn.linear_model.LogisticRegression

Evaluation: Model performance is evaluated using standard classification metrics, including:

• Accuracy
• Precision
• Recall
• F1-score
• Confusion matrices are also used to visualize the models' predictive performance for each category.

Sequence diagram illustrating the data flow and model training process: