📈 Stock Market Predictor

Bidirectional LSTM · GPU Mixed Precision · 10 Technical Indicators · 30-Day Forecast

BiLSTMMACDRSI Bollinger BandsSMA/EMA MomentumGPU float16

$205.99

Last Close (AAPL demo)

$8.24

Test RMSE

3.38%

Test MAPE

73.7%

Direction Accuracy

BiLSTM

Model

AAPL Close Price with Moving Averages & Bollinger Bands

SMA 20

SMA 50

$8.24

RMSE

3.38%

MAPE

73.7%

Direction Accuracy

BiLSTM

Architecture

Predicted vs Actual Close Price (Test Set — last 100 days)

Actual

Predicted

Model Architecture

Layer	Output Shape	Parameters
Input	(60, 10)	—
Bidirectional LSTM (64 units)	(60, 128)	38,400
BatchNormalization + Dropout(0.2)	(60, 128)	512
Bidirectional LSTM (32 units)	(64,)	41,216
BatchNormalization + Dropout(0.2)	(64,)	256
Dense(32, relu) + Dropout(0.1)	(32,)	2,080
Dense(16, relu)	(16,)	528
Dense(1, float32)	(1,)	17

30-Day Price Forecast

Historical

Forecast

Forecast Table

Project Overview

This project predicts stock closing prices using a Bidirectional LSTM neural network trained on real historical OHLCV data enriched with 10 technical indicators. Training is GPU-accelerated using TensorFlow's MirroredStrategy and float16 mixed precision.

Technical Indicators Used

✅ SMA 20 / SMA 50 (trend)

✅ EMA 12 / EMA 26 (trend)

✅ MACD + Signal (momentum)

✅ RSI 14 (overbought/oversold)

✅ Bollinger Band Width

✅ Daily Return & 20d Volatility

✅ 5-Day Price Momentum

✅ Volume Ratio (vs 20d avg)

GPU Acceleration

Mixed precision (float16) training with MirroredStrategy gives ~2–3× speedup on NVIDIA T4/P100/A100. Enable on Kaggle: Settings → Accelerator → GPU T4 x2. Enable on Colab: Runtime → Change runtime type → T4 GPU.

Dataset

Price/Volume Data for All US Stocks & ETFs — Daily OHLCV data for all NYSE/NASDAQ stocks. Trained on AAPL. 80/10/10 split.

⚠️ Disclaimer

Educational project only. Stock price predictions are inherently uncertain. Do NOT use this for real investment decisions.