Electric Vehicle Purchase Prediction

A scikit-learn Logistic Regression pipeline that estimates the probability of a customer purchasing an electric vehicle.

Created by Anjula Prasad for the Kaggle Playground Series S6E9 competition.

Training notebook and source code

Model details

Task: Binary classification
Target: Will_Buy_EV
Classes: No, Yes
Input features: 13 customer attributes
Numeric preprocessing: StandardScaler
Categorical preprocessing: OneHotEncoder
Classifier: Logistic Regression
Evaluation metric: ROC-AUC

The uploaded bundle contains the fitted preprocessing and classifier, input column names, positive-class label, training scikit-learn version, and validation ROC-AUC.

Input features

Numerical	Categorical
Age	Gender
Annual_Income_USD	City_Type
Daily_Commute_km	Current_Car_Type
Number_of_Cars_Owned	Home_Charging_Possible
Charging_Stations_Near_Home	Subsidy_Available
Charging_Stations_Near_Work	Range_Anxiety_Level
Environmental_Concern_Level

The id column is excluded from training.

Evaluation

Evaluation	ROC-AUC
Local validation — notebook output	0.93796
Kaggle public leaderboard — author reported	0.93737

The dataset contains 668,665 training rows. A stratified 80/20 split with random_state=42 produces:

Training: 534,932 customers
Validation: 133,733 customers

Preprocessing was fitted only on the training partition. The saved baseline was trained on that partition.

The leaderboard score is reported in the source repository and has not been independently verified here.

Usage

Use the training environment's dependency versions from the source repository. Install huggingface_hub if needed:

pip install huggingface_hub

Prepare a customers.csv file containing the 13 input columns above. Use the same categorical labels and numeric units as the competition data.

import joblib
import pandas as pd
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="amarakoon2001/ev-purchase-prediction",
    filename="ev_purchase_pipeline_bundle.joblib",
)

bundle = joblib.load(path)
pipeline = bundle["pipeline"]

customers = pd.read_csv("customers.csv")
X = customers[bundle["input_columns"]]

positive_index = list(pipeline.classes_).index(
    bundle["positive_class"]
)

probabilities = pipeline.predict_proba(X)[:, positive_index]

results = customers.copy()
results["EV_purchase_probability"] = probabilities
print(results)

The output is the estimated probability of class Yes. No custom classification threshold was selected for this release.

Only load Joblib files from trusted sources. Use dependency versions compatible with the training environment; the bundle records the training scikit-learn version.

Intended use and limitations

This is a competition baseline and educational portfolio project.

Performance on competition data does not establish performance for real customers, other regions, or future market conditions. ROC-AUC measures ranking performance and does not establish probability calibration.

The model has not been validated for production marketing or customer decisions. It predicts associations and does not establish causal effects.

Author

Anjula Prasad
University of Moratuwa — Information Technology

Downloads last month: -