nqmgaming/blockads-onnx

🛡️ BlockAds On-Device AI Domain Classifier (ONNX)

A lightweight, ultra-fast On-Device Machine Learning model (~900 KB) designed for mobile adblockers, DNS VPNs, and firewall applications. It evaluates domain names and detects advertisement, telemetry, and tracking infrastructure in < 0.02ms without cloud dependency.

🚀 Key Specifications

• Format: ONNX (adblock_model.onnx) & Scikit-Learn (adblock_domain_classifier.joblib)
• Model Size: 905.6 KB (Mobile ready)
• Inference Speed: > 45,000 domains / second on CPU (Single Thread)
• Trained Samples: 1,000,000 balanced domains (Tranco, Cisco Umbrella, StevenBlack, EasyPrivacy)
• Recommended Threshold: 0.75 (75%) for 0% False Positives on banking and news websites.

---

📊 Benchmark Results

Benchmark Dataset	Size	Threshold	Block Rate	Safe Sites Accuracy
Legitimate Websites & Banking	60	0.75	0.00% FP	100.00% PASS
StevenBlack Unified Hosts	10,000	0.75	35.08%	—
EasyList Privacy (EasyPrivacy)	10,000	0.75	18.79%	—
Zero-Day MMP Trackers (Adjust, AppsFlyer)	Tested	0.75	100.00%	—

Dual-Tier Architecture Note: In production mobile applications, use a Trie / Bloom Filter for known static blocklists (Tier 1) and this ONNX Model (Tier 2) to intercept novel zero-day and DGA tracking domains before they appear in public filter lists.

---

💻 Quickstart (Python + ONNX Runtime)

import numpy as np
import onnxruntime as ort
import re, math
from collections import Counter

# 1. Load Model
session = ort.InferenceSession("adblock_model.onnx")
input_name = session.get_inputs()[0].name

# 2. Extract Features (16 calibrated topological & lexical features)
def extract_features(domain: str) -> list[float]:
    # (See scripts/test_model_coverage.py for full feature extraction logic)
    pass

# 3. Predict
domains = ["app.adjust.com", "vietcombank.com.vn", "random-tracker-xyz.buzz"]
# X = np.array([extract_features(d) for d in domains], dtype=np.float32)
# probs = session.run(None, {input_name: X})[1]
# for d, p in zip(domains, probs):
#     print(f"{d}: {'BLOCK' if p.get(1, 0.0) >= 0.75 else 'PASS'} ({p.get(1, 0.0)*100:.1f}%)")

📱 Mobile Deployment

• iOS: Convert to CoreML via coremltools or use onnxruntime-c / Swift pod.
• Android: Use com.microsoft.onnxruntime:onnxruntime-android via Maven Central.