gpm123/phishing-website-screenshots

Phishing Website Screenshots

A dataset of 8,370 full-page website screenshots labelled as legitimate or phishing, intended for training and evaluating visual phishing-detection models.

Contents

Screenshots were captured at a desktop viewport (1920×1080) as PNG images.

Structure

legitimate/<brand>/<page>.png
phishing/<source>/<page>.png
metadata.csv

metadata.csv has one row per image. Key columns:

• —file_name — path to the image (relative to the repo root)
• —label — 0 = legitimate, 1 = phishing
• —label_name — legitimate / phishing
• —url, final_url — captured URL and post-redirect URL
• —brand, brand_normalized, page_type
• —image_width, image_height, file_size_bytes, image_hash
• —is_blank, is_captcha, is_error_page, status_code, timestamp

Note: 7,465 images have full metadata; the remaining 905 have label/label_name/file_name populated with other fields left blank.

Usage

from datasets import load_dataset

ds = load_dataset("shresthsamyak/phishing-website-screenshots")  # replace with your repo id
print(ds["train"][0])          # image + metadata columns

Or load the metadata directly:

import pandas as pd
from huggingface_hub import hf_hub_download

csv_path = hf_hub_download(
    "shresthsamyak/phishing-website-screenshots", "metadata.csv", repo_type="dataset"
)
df = pd.read_csv(csv_path)

Intended use & disclaimer

This dataset is provided for defensive security research and education (e.g. building phishing-detection classifiers). Screenshots are images of publicly reachable web pages captured at collection time; brand names, logos, and page content remain the property of their respective owners. The phishing samples depict malicious impersonation pages — do not reproduce or host them. Review applicable laws and third-party rights before redistribution or commercial use.

Class imbalance

The dataset is imbalanced (~18:1 legitimate:phishing). Consider class weighting, resampling, or metrics robust to imbalance (PR-AUC, F1) when training.