📈 NextPrice — Stock Price Predictor

A production-grade LSTM-based stock price forecasting system — real-time data ingestion, deep learning inference, async task queuing, REST API, and a Telegram bot interface — all containerized and CI/CD ready.

---

🚀 Try It Instantly — Interactive Colab Notebook

The Colab notebook walks through the complete ML pipeline end-to-end — no local setup required:

• 📥 Fetching 10 years of real stock data via yfinance
• ⚙️ MinMax scaling + 100-timestep sliding window construction
• 🧠 Stacked LSTM training (128 → 64 units) with Adam optimizer
• 📊 Evaluation: MSE, RMSE, R² on held-out data
• 📈 Actual vs. Predicted price visualization
• 🔮 Next-day price prediction

---

🏆 Model Performance — Real Benchmark Results

Evaluated on 10 years of daily OHLCV data (2,454 training samples per ticker) using the trained LSTM model.

Ticker	R² Score	RMSE	MSE	Next-Day Prediction
AAPL	0.9984	$3.08	9.49	$303.45
MSFT	0.9985	$5.27	27.78	$414.81
TSLA	0.9956	$9.08	82.50	$403.84

R² > 0.995 across all tickers — the model explains >99.5% of price variance on unseen trading days.

---

🧠 Machine Learning Core

Model Architecture

Input  →  (batch, 100 timesteps, 1 feature)
   │
   ▼
LSTM(128 units, return_sequences=True, activation=tanh)
   │
   ▼
LSTM(64 units, activation=tanh)
   │
   ▼
Dense(25, activation=linear)
   │
   ▼
Dense(1, activation=linear)   ← next-day price (inverse-scaled to USD)

Hyperparameter	Value
Architecture	Stacked LSTM — 2 recurrent layers
Input window	100 trading days (timesteps)
Input features	Closing price (MinMax scaled → [0, 1])
Output	Next-day closing price
Optimizer	Adam (lr=0.001, β₁=0.9, β₂=0.999)
Loss function	Mean Squared Error (MSE)
Training samples	2,454 sequences per ticker (10 years)
Model size	1.4 MB (.keras format)
Framework	TensorFlow 2.19 / Keras

ML Data Pipeline

Yahoo Finance API  →  10 years OHLCV data  (period="10y")
        │
        ▼
  Extract Closing Price  →  shape: (N, 1)
        │
        ▼
  MinMaxScaler  →  values ∈ [0, 1]
        │
        ▼
  Sliding Window (size=100)  →  X: (2454, 100, 1)  |  y: (2454, 1)
        │
        ▼
  Stacked LSTM Forward Pass
        │
        ▼
  Inverse Transform  →  price in USD
        │
        ▼
  Metrics: MSE · RMSE · R²  +  Matplotlib plots saved to /media/

---

🏗️ System Architecture

                      ┌─────────────────────────────────────┐
                      │          Client Interfaces           │
                      │  Web Dashboard │ REST API │ Telegram │
                      └──────────┬──────────┬───────────────┘
                                 │          │
                          ┌──────▼──────────▼──────┐
                          │    Django Application    │
                          │  (DRF + JWT Auth + Rate  │
                          │       Limiting)          │
                          └────────────┬────────────┘
                                       │
                    ┌──────────────────▼──────────────────┐
                    │          Celery Task Queue           │
                    │        (Redis message broker)        │
                    └──────────────────┬──────────────────┘
                                       │
                    ┌──────────────────▼──────────────────┐
                    │       StockPredictor Service         │
                    │  yfinance → preprocess → LSTM infer  │
                    │  → metrics → Matplotlib plots → DB   │
                    └─────────────────────────────────────┘

---

✨ Features

Feature	Description
🧠 LSTM Inference Engine	Stacked LSTM (128→64 units) trained on 2,454 sequences per stock
📊 R² > 0.995	Model explains >99.5% of closing price variance across AAPL, MSFT, TSLA
📈 Auto Visualization	Matplotlib charts: price history + actual vs. predicted overlay
⚡ Async Queue	Celery + Redis for non-blocking, scalable prediction jobs
🌐 REST API	JWT-authenticated endpoints for programmatic access
🤖 Telegram Bot	/predict AAPL — get ML predictions directly in chat
🔒 Rate Limiting	Built-in abuse prevention via django-ratelimit
🐳 Docker	Full containerized deployment with Docker Compose
✅ CI	Automated test suite runs on every push via GitHub Actions

---

🛠️ Tech Stack

ML / Data

Library	Version	Role
TensorFlow / Keras	2.19.0	Stacked LSTM training & inference
Scikit-learn	1.7.0	MinMaxScaler, MSE, R² metrics
NumPy	—	Sliding window sequences, array ops
yfinance	0.2.64	Real-time & historical OHLCV data
Matplotlib	3.10.3	Prediction chart generation

Backend / Infrastructure

Library	Version	Role
Django	5.0.6	Web framework
Django REST Framework	3.16.0	REST API
SimpleJWT	5.5.0	JWT authentication
Celery	5.4.0	Async task queue
Redis	5.0.1	Message broker & cache
python-telegram-bot	22.1	Telegram bot interface
Gunicorn	23.0.0	WSGI production server
Docker	—	Containerization

---

📁 Project Structure

nextprice/
├── core/
│   ├── services/
│   │   └── predictor.py        # ML pipeline: fetch → scale → window → LSTM → metrics
│   ├── management/commands/
│   │   └── predict.py          # CLI: python manage.py predict --ticker AAPL
│   ├── tasks.py                # Celery async tasks (web + Telegram delivery)
│   ├── telegram/               # Telegram bot handlers
│   ├── models.py               # Prediction model (ticker, metrics JSON, plot URLs)
│   ├── views.py                # DRF API views
│   └── tests.py                # Test suite
├── user/                       # Auth & user management
├── templates/                  # Web dashboard HTML
├── static/                     # CSS / JS assets
├── zproject/                   # Django settings & routing
├── stock_prediction_model.keras # Trained LSTM model (1.4 MB)
├── docker-compose.yml
└── requirements.txt

---

🚀 Quick Start

Option A — Docker (Recommended)

git clone https://github.com/jitendra-ky/nextprice
cd nextprice
cp .env.example .env          # edit .env with your keys
docker-compose up --build

Open http://localhost:8000/dashboard

Option B — Local Dev

git clone https://github.com/jitendra-ky/nextprice
cd nextprice
python -m venv .venv
.venv\Scripts\activate          # Windows
# source .venv/bin/activate     # macOS / Linux
pip install -r requirements.txt
cp .env.example .env
python manage.py migrate
python manage.py createsuperuser

Start services in separate terminals:

# Terminal 1 — Django
python manage.py runserver

# Terminal 2 — Celery worker (required for Telegram bot)
python manage.py startcelery

# Terminal 3 — Telegram bot
python manage.py telegrambot

Run a prediction from CLI

python manage.py predict --ticker AAPL
# → Predicted next-day price: $303.45
# → R²: 0.9984  |  RMSE: $3.08  |  MSE: 9.49

---

🔍 REST API

Method	Endpoint	Description	Auth
POST	/api/v1/register/	Register new user	No
GET	/api/v1/register/	Get user profile	JWT
POST	/api/v1/token/	Obtain JWT token	No
POST	/api/v1/token/refresh/	Refresh JWT token	No
POST	/api/v1/predict/	Run LSTM prediction	JWT
GET	/api/v1/predictions/	List user predictions	JWT
GET	/healthz/	Health check	No

Example — AAPL prediction response:

{
  "ticker": "AAPL",
  "next_day_price": 303.45,
  "mse": 9.491,
  "rmse": 3.08,
  "r2": 0.9984,
  "plot_urls": [
    "/media/plots/AAPL_history.png",
    "/media/plots/AAPL_pred_vs_actual.png"
  ]
}

---

🤖 Telegram Bot

Command	Action
/start	Register and link your account
/predict AAPL	Run LSTM prediction for ticker
/latest	View your most recent prediction
/help	Show all commands

---

🧪 Testing & CI

python manage.py test          # Run full test suite
ruff check .                   # Linting

Automated tests run on every push via GitHub Actions (see .github/workflows/django.yml).

---

📄 License

MIT License — see LICENSE for details.