Mobile Price Classification Notebook

Author: J. Wong
Date: 2026-02-09

This repository contains a Jupyter Notebook that demonstrates a complete machine learning workflow on the Mobile Price Classification dataset. The goal is to predict the price_range of mobile devices using multiple classification algorithms while balancing performance and interpretability.

What’s Included

• Data Import & Preprocessing
  • Dataset inspection and cleaning
  • Feature/target correlation analysis
  • Multicollinearity checks
• Exploratory Data Analysis
  • Visualizations of feature distributions
  • Identification of most predictive features
• Regression & Classification Modeling
  • Logistic Regression (L1/L2, multinomial)
  • K-Nearest Neighbors (KNN) and elbow curve analysis
  • Support Vector Machines (Linear SVC, RBF SVC, GridSearch tuning)
  • Decision Trees (with and without GridSearch)
  • Ensemble methods:
    • Bagging
    • Random Forests (with GridSearch)
    • Extra Trees
    • Gradient Boosting, AdaBoost
    • Voting & Stacking Classifiers
    • XGBoost (including eval metrics & boosting iterations analysis)
• Hyperparameter Tuning & Evaluation
  • GridSearchCV
  • Cross-validation
  • Train/test splits
  • Metrics: Accuracy, Log Loss, Error Rate, ROC-AUC, Confusion Matrices
• Model Interpretability
  • Feature importance & permutation importance
  • Global surrogate models
  • LIME for local explanations

Key Results

• Ensemble and boosting methods delivered highest predictive performance.
• Features like ram, battery_power, px_height, and px_width were consistently most influential.
• Models are both accurate and explainable, demonstrating a strong ML workflow.

Dataset

• Mobile Price Classification on Kaggle by A. Sharma
• Target variable: price_range (categorical price levels for mobile devices)

Usage

• Open Mobile_Price_Classification.ipynb in Jupyter Notebook or Lab.
• Run cells sequentially to reproduce preprocessing, modeling, evaluation, and interpretation steps.
• Modify hyperparameters or models to experiment further.