The Titanic StreamLit Website is an interactive web platform showcasing machine learning models developed for the Kaggle Titanic dataset. The website features a homepage and dedicated pages for Neural Network, Random Forest, and Gradient Boosted Trees models. This project serves as a testament to the deployment of machine learning models to the cloud, leveraging Streamlit for seamless integration.
-
Home Page: The homepage (
index.html
) introduces the project, providing a glimpse into the machine learning models used and the data preprocessing techniques. Links to GitHub repositories and other model pages are also available. -
Model Pages:
- Neural Network (
Neural_Network.py
): This page, structured with HTML (Neural_Network_1.html
andNeural_Network_2.html
), presents the neural network model. Users can input custom values like age and gender to receive survival predictions. - Random Forest (
Random_Forest.py
): Similar in structure to the Neural Network page, this page (Random_Forest_1.html
andRandom_Forest_2.html
) focuses on the Random Forest model, allowing users to interact and receive predictions. - Gradient Boosted Trees (
Gradient_Boosted.py
): Showcasing the Gradient Boosted Trees model, this page (Gradient_Boosted_1.html
andGradient_Boosted_2.html
) also allows for interactive user input and predictions.
- Neural Network (
- Data Preprocessing: Separate preprocessing scripts for each model (
data_preprocessing_neural_network.py
,data_preprocessing_Random_Forest.py
,data_preprocessing_Gradient_Boosted.py
) ensure data is formatted correctly for model input. - Model Deployment: Models are saved and deployed using Streamlit. For instance,
model_12_saved.h5
represents a saved neural network model, andmodel3_GB/saved_model.pb
for the Gradient Boosted Trees. - Requirements: The
requirements.txt
file lists all the necessary libraries and dependencies required to run the models and the website.
- Model Training: Detailed training and fine-tuning of models are evident in the code and HTML descriptions, ensuring optimal performance.
- Evaluation and Scoring: The project boasts a high score of 80.622 in the Kaggle competition, indicating the efficacy of the models and the engineering pipeline.
- This project illustrates the integration of HTML with Streamlit, displaying the capability to create interactive web applications for machine learning models.
- It highlights the deployment of models to the cloud, allowing for real-time interaction and prediction based on user input.
- The repository serves as an educational tool for understanding model deployment, cloud integration, and the use of Streamlit in machine learning projects.
For an in-depth view and understanding of this project, visit the GitHub Repository and explore the various HTML and Python files that constitute the website's structure and functionality.