ESGeePeeTee
ESGeePeeTee is an automated ESG score predictor powered by machine learning. ESG scoring is notoriously known to be laborious, expensive, and prone to human bias. This solution aims to perform an objective and cost-efficient ESG evaluation of the company of your choice. Developed for the Nomura Trailblazer AI Hackathon 2024 and achieved 2nd place.
* ESG: Environment, Social, Governance scores are used for company screening before investments.
💯 S&P Scoring Methodoloy
The tool is built around the reputable S&P scoring methodology, which is widely adopted across industries. Each company is categorized by industry, and scores are calculated across the 3 main dimensions.
📊 Data Source
To automate data collection and minimize bias, the system utilizes the GDELT project - a massive, real-time database of global news and events. Â GDELT's Global Knowledge Graph (GKG) provides trillions of datapoints, updated every 15 minutes, and includes advanced features such as sentiment analysis and thematic summarization. By querying GDELT with custom search terms generated through LLMs, ESGeePeeTee efficiently retrieves all relevant news articles for each ESG criterion.
🤖 Machine Learning Model
The core of the scoring engine is powered by AutoGluon, an AutoML library that employs multi-layer, stack ensembling techniques. This approach combines the predictive strengths of multiple machine learning models to produce highly accrurate ESG scores (8.97 RMSE).
🌟 Key Features 🌟
- Fully automated ESG scoring based on the S&P Global Methodology
- Real-time data integration from GDELT's global news repository
- Multi-model ensemble learning for robust and accurate predictions
- Delivers comprehensive ESG reports in minutes, free from manual bias