Shopping Predictor

Project Description

This project involves building a nearest-neighbor classifier to predict whether online shopping customers will complete a purchase based on their browsing behavior. The classifier uses a dataset of around 12,000 user sessions, analyzing features like pages visited, session duration, bounce rates, and more.

Features

• Data Preprocessing: Load and process data from a CSV file.
• Model Training: Train a k-nearest-neighbor classifier using scikit-learn.
• Evaluation Metrics: Calculate sensitivity (true positive rate) and specificity (true negative rate).

Usage

1. Setup:

   pip install scikit-learn

2. Run the Program:

   python shopping.py shopping.csv

Functions to Implement

1. load_data(filename): Load and preprocess data from CSV.
2. train_model(evidence, labels): Train the k-nearest-neighbor model.
3. evaluate(labels, predictions): Calculate sensitivity and specificity.

Evaluation

• Correct: Number of correct predictions.
• Incorrect: Number of incorrect predictions.
• True Positive Rate: Proportion of actual positives correctly identified.
• True Negative Rate: Proportion of actual negatives correctly identified.

Results Explanation

Run the program:

python shopping.py shopping.csv

After running the program, the output will include:

Correct: 4078
Incorrect: 854
True Positive Rate: 40.61%
True Negative Rate: 90.24%

Interpretation of Results

• Correct Predictions (4078): The number of user sessions where the model accurately predicted whether a purchase was completed.
• Incorrect Predictions (854): The number of user sessions where the model's prediction was incorrect.
• True Positive Rate (Sensitivity) (40.61%): This metric indicates that 40.61% of the actual purchases were correctly identified by the model. It reflects the model's ability to identify users who will make a purchase.
• True Negative Rate (Specificity) (90.24%): This metric indicates that 90.24% of the non-purchases were correctly identified by the model. It reflects the model's ability to identify users who will not make a purchase.

These metrics help in understanding the performance of the classifier in distinguishing between customers who are likely to complete a purchase and those who are not. The higher the sensitivity and specificity, the better the model is at making accurate predictions.