EV Range Predictor

A machine learning model that predicts the estimated driving range (in km) of an electric vehicle based on its technical specifications. Built with a Random Forest regressor and served via an interactive Streamlit UI.

---

Overview

Range anxiety is one of the biggest barriers to EV adoption. This project trains a regression model on a dataset of ~478 real-world EVs from the EV Database, covering 59 brands and a wide variety of vehicle segments, body types, and drivetrains. Given a vehicle's specs, the app instantly predicts its range and classifies it as short, mid, or long range.

---

Project Structure

├── ev_data.csv              
├── ev_range_predictor.ipynb   
├── random_forest_model.pkl    
├── columns.pkl               
├── target_encoder.pkl         
└── app.py                     

---

Notebook Walkthrough

1. Exploratory Data Analysis

• Inspected feature distributions and value counts for categorical columns (brand, segment, body type, drivetrain)
• Identified and handled missing values: number_of_cells filled with the column mean; remaining rows with nulls dropped
• cargo_volume_l coerced to numeric, nulls filled with the median

2. Feature Engineering

• Dropped: source_url, model, battery_type, fast_charge_port
• One-hot encoding: car_body_type, segment, drivetrain
• Target encoding: brand — encoded using category_encoders.TargetEncoder with smoothing=10 to avoid overfitting on rare brands

3. Modelling

Two regressors were trained and compared on an 80/20 split:

Model	MAE (km)	R²
Random Forest	11.90	0.9733
XGBoost	12.31	0.9750

Random Forest was selected as the final model and saved as random_forest_model.pkl.

---

Streamlit App

The app provides a form-based interface grouped into three sections:

Vehicle Identity

• Brand (59 options), Body Type, Drivetrain, Segment

Battery & Performance

• Battery Capacity (kWh), Fast Charging Power (kW DC), Top Speed, 0–100 km/h acceleration, Torque, Efficiency (Wh/km)

Dimensions & Capacity

• Length, Width, Height, Seats, Cargo Volume, Towing Capacity, Number of Cells

On submission, the app returns:

• Predicted range in km and miles
• A contextual label: Short range (< 300 km) / Mid range (300–500 km) / Long range (> 500 km)

---

Tech Stack

Layer	Library
Data manipulation	pandas, numpy
Visualization	matplotlib, seaborn
Encoding	category_encoders (TargetEncoder)
Machine learning	scikit-learn (Random Forest), xgboost
Model serialization	joblib
Web UI	streamlit

---

Dataset

EV dataset downloaded from kaggle with multiple features. Also it is available to see in the repository.

---