MBTA Delay Prediction Tool

Overview

This project predicts MBTA subway delays using multiple machine learning approaches for two tasks:

1. Delay Probability Prediction
   Predict whether a train will be delayed.

2. Delay Duration Prediction
   Predict how long a delay will last.

The project compares four models:

• Long Short-Term Memory (LSTM) - prediction
• Hidden Markov Model (HMM) - prediction
• Random Forest Regressor - duration
• Gradient Boosting Regressor - duration

The system includes data preprocessing pipelines, model training, saved model artifacts, and a Streamlit-based frontend for generating predictions.

---

Goal:
Evaluate how different machine learning models perform in predicting both the likelihood and duration of MBTA delays using a limited but well-engineered set of features.

Project Structure

src/models/lstm.py
src/models/hmm.py
src/models/random_forest.py
src/models/gradient_boosting.py

src/preprocessing/
src/visualizations/

Datasets/
src/models/model_storage

app.py
requirements.txt
README.md

---

Installation

Prerequisites

Python 3.10+ (we used Python 3.14)

Required libraries:

• torch
• numpy
• pandas
• scikit-learn
• matplotlib
• streamlit
• pyyaml

---

Install Dependencies

Clone the repository:

git clone https://github.com/ColinCarnish/CS4100_Final.git
cd CS4100_Final

Install packages:

pip install -r requirements.txt

---

Running the Project

Train Models

python CS4100_Final/src/models/lstm.py  
python CS4100_Final/src/models/hmm_model.py   
python CS4100_Final/src/models/GBM.py  
python CS4100_Final/src/models/forest/forest.py                 

---

Launch Frontend

streamlit run app/main.py

---