product-image-feature-analyzer

Product Image Feature Analyzer

Isolation Forest anomaly detector + KMeans clustering over the 100-dim PCA image features of a 75,000-product image dataset.

Model Characteristics

• Type: Unsupervised (IsolationForest + KMeans)
• Inputs: 100 normalized PCA image features (img_pca_0..img_pca_99)
• Outputs: anomaly flag (1/-1) and cluster assignment (0-7)
• Anomalies found: 2% of rows flagged

Data Overview

• Dataset: anujbolewar/product-image-features (75,000 rows x 102 cols)
• Built from the user's own product image feature extraction pipeline

Evaluation Results

• IsolationForest contamination set to 2%; 1500 anomalies detected
• KMeans with k=8; silhouette on 10% subsample reported

Files

• scaler.joblib — StandardScaler
• anomaly_detector.joblib — trained IsolationForest
• cluster_model.joblib — trained KMeans (8 clusters)
• feature_cols.npy — PCA feature column names
• cluster_centers.npy — cluster centroids

Usage

import joblib
scaler = joblib.load("scaler.joblib")
anomaly_detector = joblib.load("anomaly_detector.joblib")
cluster_model = joblib.load("cluster_model.joblib")

Provenance

Trained by Anuj Bolewar (anujbolewar on Kaggle) on the product-image-features dataset.