Skip to content

Dual Bidirectional LSTM approach for time series analysis, featuring both regression (next 10 minutes prediction) and classification (direction forecasting) models. Built with TensorFlow/Keras, this implementation provides precise numerical forecasts and actionable directional insights from the Nvidia stock dataset.

Notifications You must be signed in to change notification settings

Vd1299/NVIDIA_Stock_Prices_LSTM_network_Analysis

Repository files navigation

Time Series Forecasting for Nividia Stock Price with Bidirectional LSTMs

This repository contains a dual-model approach to time series analysis and prediction using Bidirectional Long Short-Term Memory (LSTM) neural networks implemented in TensorFlow/Keras.

Project Overview

This project implements two complementary deep learning models for time series analysis:

Regression Model: Predicts continuous values for the next 10 time steps (minutes) Classification Model: Categorizes future movement direction into three classes (up, same, down)

Models leverage LSTM and bidirectional LSTM architectures to capture temporal patterns from historical time series data, providing both precise numerical forecasts and actionable directional insights. Features Bidirectional LSTM networks for capturing patterns in both forward and backward directions Dual prediction approach (regression + classification) Dropout layers for regularization to prevent overfitting Early stopping and adaptive learning rate reduction during training Comprehensive evaluation metrics for model performance assessment

Evaluation

The regression model is evaluated using Mean Squared Error (MSE) and Mean Absolute Error (MAE), while the classification model is assessed using accuracy, precision, recall, and F1-score metrics.

Data Preparation

Data is extracted from Polygon.io using the api for past 180 days at minute level interval for Nvidia stock. Time series data is preprocessed using sliding windows to create input-output pairs for training. For the classification model, continuous targets are converted to categorical labels (up, same, down) based on directional movement.

Medium Article

Article

License

MIT

About

Dual Bidirectional LSTM approach for time series analysis, featuring both regression (next 10 minutes prediction) and classification (direction forecasting) models. Built with TensorFlow/Keras, this implementation provides precise numerical forecasts and actionable directional insights from the Nvidia stock dataset.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published