Ensemble-Based Machine Learning Approach for Accurate and Cost-Effective Early Detection of Harmful Algal Blooms
This repository contains the official implementation of the research paper Ensemble-Based Machine Learning Approach for Accurate and Cost-Effective Early Detection of Harmful Algal Blooms. The approach utilizes ensemble learning, combined with data augmentation and multi-agent systems, to enhance the early detection of harmful algal blooms (HABs).
Harmful Algal Blooms (HABs) present significant ecological and economic challenges, costing the U.S. $50 million annually while impacting public health and degrading water quality. Traditional detection methods, such as manual sampling, analysis, satellite monitoring, and sensing, are time-consuming, expensive, and lack real-time monitoring capabilities. This study introduces an ensemble-based machine learning approach to predict corrected Chlorophyll-a concentration, a key indicator of HABs. Initial tests of our ensemble model revealed higher accuracy in predicting lower Chlorophyll-a concentrations compared to higher ones. To address this, we utilized large language models (LLMs) to generate synthetic data for high-value cases, effectively oversampling the long-tail data. This data augmentation resulted in a 4.77% reduction in prediction error for our ensemble model compared to training on the original dataset alone. Moreover, our final model achieved a notable 66.10% reduction in RMSE compared to conventional models using satellite data. This approach provides a scalable, cost-effective solution for early HAB detection, enhancing AI-driven environmental monitoring and prediction systems.
To install the required dependencies, run:
pip install -r requirements.txt
Before processing the data, ensure that the output folder and its contents are deleted if they exist, as the preprocess.py script will generate new files and the output folder.
To process the data, run this command:
python preprocess.py
This project includes training four models: Random Forests, Gradient Boosting, Neural Networks, and a Stacked/Ensemble model. To train all models at once, run:
python train.py
We provide multiple scripts to evaluate the models' performance on different metrics:
To evaluate the Root Mean Squared Error (RMSE) of the models, run:
python loss_RMSE.py
To evaluate the residuals of the models' predictions, run:
python residuals.py
To evaluate the models' percent error, run:
python percent_error.py
You can download pre-trained models from the following link:
We have developed a simple web application to demonstrate the functionality and use of our model for HAB prediction/detection. You can access it here: Web-app demo.
The web application is built using Streamlit. The code for the app is in the following folder: Streamlit App Code.
We welcome contributions! To contribute:
- Fork the repository.
- Clone your fork:
git clone https://github.com/your-username/Ensemble-ML-for-HABs-Detection.git - Create a new branch:
git checkout -b feature-or-bugfix-name - Make your changes and commit:
git commit -m "Description of changes" - Push to your fork:
git push origin feature-or-bugfix-name - Open a Pull Request.
This project is licensed under the MIT License. See the LICENSE file for details.