📈 Day Trade AI — Downside-Protected Stock Decision System

Loss-Avoidance Machine Learning Classifier for Retail Stock Day-Trading (NTT 9432 Case Study).

Overview • Features • Code Architecture • System Flow • Structure • Quick Start • License

---

📌 Overview

A machine learning decision support system engineered specifically for working professionals who trade Japanese equities (NTT 9432). Rather than maximizing raw trade frequency, the model is tuned with asymmetric loss penalties to skip high-risk market days and generate an actionable binary buy/skip signal at 8:30 AM before market open (Market Open Buy & Market Close Sell).

---

✨ Features (Key Outcomes & Capabilities)

Icon	Feature	Outcome & Real Proof
🛡️	Asymmetric Downside Protection	Skips negative-expectation trading days to preserve capital
⏰	8:30 AM Actionable Signal	Deterministic trade recommendation before market open for working professionals
📊	Comprehensive Technical Features	RSI, Moving Average Deviations, Volume Spikes, and ATR Volatility
📈	Realistic Backtest Verification	Evaluates Profit Factor, Maximum Drawdown, and Sharpe Ratio on historical NTT data

---

🔬 Code Architecture & Implementation

🔬 Code Implementation (stock_prediction.ipynb)

• Target Definition: $y = 1$ if $\text{Close}_t > \text{Open}_t$ (positive intraday return), else $0$.
• Feature Engineering:
  • Price Momentum: 5-day / 20-day SMA deviation, 14-day RSI, Bollinger Bands %B.
  • Volatility & Volume: Intraday ATR (Average True Range), volume spike ratios.
• Model Tuning: LightGBM Binary Classifier optimized for Precision and Profit Factor via probability threshold tuning ($P \ge 0.65$).
• Backtesting Engine: Simulates realistic execution with commissions and slippage.

---

📊 System Flow

graph TD
  CSV[(📊 NTT 9432 Historical OHLCV)] --> Feat[⚙️ Technical Indicators: RSI, SMA, %B, ATR]
  Feat --> Model[🧠 LightGBM Classifier]
  Model --> Prob[🎯 Prediction Probability P(Up)]
  Prob --> Decision{P >= 0.65 Threshold}
  Decision -->|Yes| Buy[🟢 BUY at 9:00 Open -> SELL at 15:00 Close]
  Decision -->|No| Skip[🔴 SKIP / Hold Cash]

  classDef primary fill:#f59e0b,stroke:#d97706,stroke-width:2px,color:#fff;
  classDef green fill:#10b981,stroke:#059669,stroke-width:2px,color:#fff;
  classDef red fill:#ef4444,stroke:#dc2626,stroke-width:2px,color:#fff;
  class Model,Prob primary;
  class Buy green;
  class Skip red;

Loading

---

📁 Project Structure

day_trade_ai/
├── 📁 assets/                 # High-resolution SVG banners
│   └── 🎨 hero.svg
├── 📄 stock_prediction.ipynb   # Complete data prep, model training & backtest
├── 📄 stock_price.csv         # NTT historical price dataset
├── 📄 slide.pdf               # Project presentation deck
└── 📄 README.md               # Project documentation

---

🚀 Quick Start

# 1. Clone repository
git clone https://github.com/LoNebula/day_trade_ai.git
cd day_trade_ai

# 2. Install dependencies
pip install -r requirements.txt

# 3. Launch Jupyter Notebook
jupyter notebook stock_prediction.ipynb

---

Released under the MIT License. Crafted with precision by LoNebula