Insurance Price Predictor — Linear Regression (beginner project) 🚑💡

🌐 Live Deployment

🔗 https://insurancecostpredictorby.akashchaudhari.in

Short overview

This project shows a complete, beginner-friendly workflow for predicting insurance charges using a Linear Regression model. The InsurenceHypothesis.ipynb notebook explains each step slowly: EDA, preprocessing (encoding + log transform), model training, evaluation, and model saving.

---

What’s included 🔍

• InsurenceHypothesis.ipynb — notebook with step-by-step explanations and plots.
• Data/InsuranceData.csv — dataset used in the notebook.
• insurance_model.pkl and model_columns.pkl — example saved model and column list (created by the notebook).

---

Key concepts you’ll learn 🎯

• Encoding categorical features (sex, smoker, region) appropriately.
• Handling skewed targets with log1p and converting predictions back with expm1.
• Train/test split, model training, evaluation (R², MSE, MAE), and residual analysis.
• Saving and loading a trained model with pickle for simple deployment/testing.

---

Quick start — run the notebook ✅

1. (Optional) Create and activate a Python virtual environment:

   python -m venv .venv
   & .venv\Scripts\Activate.ps1

2. Install required packages:

   pip install pandas numpy matplotlib seaborn scikit-learn

3. Open and run InsurenceHypothesis.ipynb in VS Code / Jupyter. Run cells top-to-bottom and read the markdown explanations.

---

Example — load saved model and predict (Python)

import pickle
import numpy as np

model = pickle.load(open('insurance_model.pkl', 'rb'))
model_cols = pickle.load(open('model_columns.pkl', 'rb'))

# create an input row using the saved column order
x = np.zeros(len(model_cols))
# set features: age, sex (0=m), bmi, children, smoker (0=no), region_southeast=1, etc.
# example: 30-year male, bmi=25, 1 child, non-smoker, southeast
# set indices according to model_columns.pkl and then:
# predicted_log = model.predict([x])
# predicted_charge = np.expm1(predicted_log)

---

Next steps / experiments 🚀

• Try Ridge/Lasso regularization (reduce overfitting).
• Add interaction terms (age*bmi) or polynomial features for non-linearity.
• Compare with tree-based models (RandomForest, XGBoost).
• Calibrate error: report MAE in dollars for clearer interpretation.

---

Notes for learners ✏️

• The notebook stores predictions on the log scale — always convert back with np.expm1 before interpreting dollar values.
• Read each markdown cell in InsurenceHypothesis.ipynb — the notebook intentionally explains changes step-by-step.

---

Want me to add a short requirements.txt, improve model performance (step-by-step), or create a small demo script that loads the model and predicts for user input? Reply which task you want next.