Proposing an Integrated Approach to Analyzing ESG Data via Machine Learning and Deep Learning Algorithms

Ook Lee, Hanseon Joo, Hayoung Choi, Minjong CheonSustainability 2022 · DOI

TL;DR

This paper proposes an integrated approach to analyzing ESG (Environmental, Social, and Governance) data using machine learning and deep learning. It addresses the challenge of limited ESG data by leveraging NLP for social and governance aspects and introduces methods for predicting ESG rankings and detecting adversarial attacks, offering practical tools for businesses.

What Problem Does This Paper Address?

The increasing importance of ESG factors in investment decisions and corporate strategy, particularly highlighted during the COVID-19 pandemic, has created a demand for robust ESG analysis. However, a significant challenge exists in collecting and standardizing ESG-related datasets, especially for social and governance aspects, which are often qualitative and unstructured. Existing research has explored various AI techniques for ESG analysis, but there's a gap in integrated approaches that cover data collection, analysis, adversarial attack detection, and practical ESG score prediction, especially for smaller businesses. This paper aims to bridge this gap by proposing a multi-faceted AI framework for ESG data analysis.

Methods

The paper presents five distinct experiments, each employing different AI techniques to analyze ESG data:

  • Classification: Used to determine if an investment is "good" or "bad" based on financial ratios and ESG rankings.
  • Anomaly Detection: Investigates the impact of adversarial attacks (specifically poisoning attacks) on ESG data and proposes methods to detect noise using Local Outlier Factors (LOF) and Isolation Forest (IF).
  • Regression: Predicts financial performance (fund's annual return) using ESG indicators.
  • Natural Language Processing (NLP) for Text Classification: Leverages deep learning models (GRU, LSTM, Bi-LSTM) to classify news articles into Environmental (E), Social (S), and Governance (G) categories.
  • ESG Score Prediction: Develops a straightforward method for small businesses to predict ESG scores using financial data.

The core AI techniques employed include:

  • Machine Learning: Decision Tree, Logistic Regression, Gradient Boosting, Adaptive Boosting, Random Forest, XGBoost, Light Gradient Boosting Machine (LGBM), Extra Trees, Isolation Forest.
  • Deep Learning: Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM).
  • NLP Techniques: Tokenization, lemmatization, Stop Words removal, embeddings.

Datasets

The paper utilizes several publicly available datasets:

  • Kaggle Dataset (Experiment #1 & #5): A dataset of NYSE random investments with financial ratios and an "ESG rating" row. This dataset was pre-processed using MinMax scaler and null value imputation.
  • Kaggle Dataset (Experiment #3): Data from fossilfreefunds.org, focusing on ESG indicators and fund performance. This dataset was pre-processed by removing non-ESG related indicators and handling missing values, followed by a train-test split and Robust scaling.
  • Korean News Articles (Experiment #4): News titles and articles crawled from the Korean website "Naver." This dataset was processed using tokenization, lemmatization, and Stop Words removal.

No new benchmarks were introduced; instead, existing datasets were leveraged and pre-processed for specific experimental goals.

Key Insights and Ideas

  • Integrated AI for ESG: The paper demonstrates the power of combining various AI techniques (ML, DL, NLP) to tackle different facets of ESG data analysis, from raw data to actionable insights.
  • NLP for Social & Governance Data: Effectively using NLP to extract E, S, and G labels from text data addresses a critical challenge in ESG analysis where structured data is scarce.
  • Adversarial Attack Awareness: Highlighting the vulnerability of ESG datasets to adversarial attacks and proposing detection methods is crucial for ensuring the reliability of AI-driven ESG analysis.
  • Democratizing ESG Prediction: The proposed simple method for predicting ESG scores using financial data offers a valuable tool for small businesses that may lack the resources for complex ESG assessments.
  • Financial Indicators as ESG Proxies: The finding that financial variables can effectively predict ESG scores suggests a strong underlying correlation that can be exploited for practical applications.

Key Results

  • Experiment #1 (Investment Classification): Achieved high accuracy, with LGBM, Extra Trees, Gradient Boosting, and Decision Tree models exceeding 99% accuracy.
  • Experiment #2 (Noise Detection): Demonstrated that noise significantly impacts classifier accuracy (e.g., LGBM dropped from 99.5% to 88.5%). Anomaly detection models like LOF and Isolation Forest showed high accuracy (95.156% and 84.1% respectively) in detecting noise.
  • Experiment #3 (Financial Performance Prediction): Extra Tree regression achieved the best Mean Absolute Error (MAE) of 0.5849, indicating that ESG indicators can predict fund returns with a small margin of error.
  • Experiment #4 (News Classification): Achieved approximately 90% accuracy in classifying news articles into E, S, and G labels using embeddings and Bi-LSTM.
  • Experiment #5 (ESG Score Prediction): Gradient Boosting and XGBoost yielded the most accurate predictions with low RMSE and MAE scores (e.g., 1.18 RMSE and 1.4 MAE).

Limitations and Future Work

  • Dataset Limitations: The authors acknowledge that the datasets used were limited, particularly in the fourth experiment where data was self-crawled. They suggest that cooperating with specific companies to gather abundant, real-world data would be beneficial.
  • Language Specificity: The NLP experiment for news classification was conducted primarily in Korean. Future work includes expanding this to other languages like English, French, and Spanish to broaden applicability.
  • Real-World Application: While the results are promising, the paper suggests that further research should focus on conducting real-life analyses with more extensive datasets and collaborations with firms to validate the practical utility of the proposed methods.

Summary updated

Abstract

In the COVID-19 era, people face situations that they have never experienced before, which alerted the importance of the ESG. Investors also consider ESG indexes as an essential factor for their investments, and some research yielded that the return on sustainable funds is more significant than on non-sustainable ones. Nevertheless, a deficiency in research exists about analyzing ESG through artificial intelligence algorithms due to adversity in collecting ESG-related datasets. Therefore, this paper suggests integrated AI approaches to the ESG datasets with the five different experiments. We also focus on analyzing the governance and social datasets through NLP algorithms and propose a straightforward method for predicting a specific firm’s ESG rankings. Results were evaluated through accuracy score, RMSE, and MAE, and every experiment conducted relevant scores that achieved our aim. From the results, it could be concluded that this paper successfully analyzes ESG data with various algorithms. Unlike previous related research, this paper also emphasizes the importance of the adversarial attacks on the ESG datasets and suggests methods to detect them effectively. Furthermore, this paper proposes a simple way to predict ESG rankings, which would be helpful for small businesses. Even though it is our limitation that we only use restricted datasets, our research proposes the possibility of applying the AI algorithms to the ESG datasets in an integrated approach.
BibTeX
@inproceedings{lee2022proposing,
  author = {Ook Lee and Hanseon Joo and Hayoung Choi and Minjong Cheon},
  title = {{Proposing an Integrated Approach to Analyzing ESG Data via Machine Learning and Deep Learning Algorithms}},
  booktitle = {Sustainability},
  year = {2022},
  doi = {10.3390/su14148745},
}
Figure