| # CryptoTrendPredictor_LSTM | |
| ## 📈 Overview | |
| This repository hosts a custom **Long Short-Term Memory (LSTM)** recurrent neural network designed for **single-step time-series prediction** of cryptocurrency prices. Specifically, this model is trained to predict the **normalized closing price of Bitcoin (BTC)** for the next day, utilizing a diverse set of technical indicators and on-chain metrics. | |
| ## 🧠 Model Architecture | |
| The model implements a stacked LSTM architecture, which is highly effective for capturing long-term dependencies in sequential data like price time series. | |
| * **Architecture:** Stacked LSTM (3 layers) with a final Dense layer for regression. | |
| * **Input Features (`input_size=10`):** 10 features including price, volume, technical indicators (RSI, MACD), market correlation, and external factors (Google Trends, Social Media Sentiment). | |
| * **Sequence Length (`sequence_length=60`):** The model takes the past **60 time steps (days)** of feature data as input to make one prediction. | |
| * **Hidden State Size (`hidden_size=128`):** 128 units per LSTM layer. | |
| * **Output:** A single floating-point value representing the normalized predicted closing price for the next time step. | |
| * **Loss Function:** Mean Squared Error (MSE). | |
| ## 💡 Intended Use | |
| * **Price Forecasting:** Providing a **probabilistic forecast** of the next-day price movement for risk assessment. | |
| * **Signal Generation:** Generating buy/sell/hold signals when the predicted value crosses predefined thresholds. | |
| * **Comparative Analysis:** Benchmarking against simpler linear or technical analysis models. | |
| ### Data Requirements for Inference | |
| The model expects a numpy array/tensor of shape `(batch_size, sequence_length, input_size)`. The input features **must** be normalized using the same min/max scaling applied during training. | |
| ### How to use (Conceptual Example) | |
| ```python | |
| import numpy as np | |
| # Assuming the model is loaded from a framework like PyTorch or TensorFlow | |
| # and inference data is ready and pre-processed. | |
| # Replace with actual data (60 days of 10 features, normalized) | |
| input_data = np.random.rand(1, 60, 10).astype(np.float32) | |
| # prediction = model(input_data) | |
| # print(f"Normalized Prediction: {prediction.item()}") | |
| # Inverse transform the prediction to get the actual price forecast | |
| # forecasted_price = scaler.inverse_transform(prediction) | |
| # print(f"Forecasted BTC Price: ${forecasted_price.item():.2f}") |