The 17 UN Sustainable Development Goals: Classification of Research Topics Using BERT and Logistic Regression

An academic institution with over 200 lecturers has produced more than 3,000 research articles between 2018 and 2023. Accurately classifying these research outputs according to the 17 United Nations Sustainable Development Goals (UN SDGs)—a global agenda addressing issues such as poverty, education, gender equality, clean energy, and climate action—is vital for demonstrating institutional contributions to sustainability and supporting faculty accreditation processes. Traditionally, the Research and Community Service Institute of private universities has performed this classification manually, which is inefficient and time-consuming. To address this challenge, two machine learning-based text classification systems were developed and evaluated. The model was trained on a dataset of 76,958 records. The first approach implements a Bidirectional Encoder Representations from Transformers (BERT) model, a state-of-the-art deep learning framework in Natural Language Processing. Preprocessing was performed using NLTK, and the model was fine-tuned over 4 epochs with a learning rate of 2e-5 and a batch size of 32, using a 70/30 train-test split. This model delivered superior performance, with an accuracy of 90.68%, precision of 0.99, recall of 0.82, and an F1-score of 0.87. The second approach utilizes a Logistic Regression model with TF-IDF (Term Frequency-Inverse Document Frequency) for text vectorization. This model employs the L1 penalty and the Saga solver, trained with 80% of the dataset and tested on the remaining 20%, without additional data cleaning. It achieved an accuracy of 90.01%, a precision of 0.86, recall of 0.82, and an F1-score of 0.84. Both models demonstrated strong performance, but the BERT-based model provided better precision and overall classification quality. The findings show that both models deliver strong classification performance, with the BERT-based model providing superior precision and overall quality. These systems have been presented to the university for potential adoption, offering a more efficient and consistent approach to aligning institutional research with 17 UN SDGs.