ClickStream Classifier (Demo)

Tiny full-stack demo that generates click events, classifies them as human or bot, stores them in Postgres and visualizes results with Chart.js (daily Real/Fake counts, Fake Rate, and a Users table).

Stack

• Node 18+, Express, CORS
• PostgreSQL (pg)
• Chart.js (frontend)
• Vanilla HTML/CSS/JS

Project structure

repo/

    backend/

      server.js

      db.js

      package.json

      .env.example

    public/

      index.html

      app.js

      styles.css

    README.md

Prereqs

• Node 18+
• PostgreSQL running locally
• VS Code Live Server (or any static file server)

Environment

Create backend/.env (copy from .env.example)

DATABASE_URL=postgres://bytebrew_user:bytebrew_pass@localhost:5432/bytebrew

PORT=3000

Database schema

CREATE TABLE IF NOT EXISTS public.events (
  id SERIAL PRIMARY KEY,
  ad_id TEXT NOT NULL,
  clicks JSONB NOT NULL,
  classification TEXT NOT NULL,
  domain TEXT,
  user_agent TEXT,
  browser TEXT,
  ip TEXT,
  country TEXT,
  created_at TIMESTAMPTZ DEFAULT now()
);

CREATE INDEX IF NOT EXISTS events_created_at_idx ON public.events (created_at);
CREATE INDEX IF NOT EXISTS events_classification_idx ON public.events (classification);

Run

1. Install & start backend API

   cd backend

   npm install

   npm run dev

   (Health check: http://127.0.0.1:3000/api/events should return JSON)

2. Serve frontend with Live Server

   CORS is enabled on the API so cross-origin calls from :5500 will work

3. ENDPOINTS

   • POST /api/generate?pattern=human|bot creates & classifies a synthetic event

   • GET /api/events returns latest 100 events (for table + totals donut)

   • GET /api/metrics daily aggregates:

     [
       { "day":"YYYY-MM-DD", "human":N, "bot":M, "total":T, "fakeRate": M/T }
     ]
     

Features delivered

Charts: Real/day, Fake/day, Fake Rate/day (Chart.js)

Totals donut (human vs bot, all recent events)

Users table: adId, domain, userAgent, browser, ip, country, classification, createdAt, clickCount

Randomized generator: variable count & timing for human/bot patterns

Lightweight classifier: interval stats + heuristics

Debug overlay (toggle) for raw event JSON