Machine Learning‐Based Automated Classification of SDG Alignment in Indian Sustainability Startups

Addressing global sustainability challenges demands innovative, agile, and data‐driven approaches, with startups increasingly recognised as pivotal actors within dynamic sustainability ecosystems, directly contributing to advancing the United Nations Sustainable Development Goals (SDGs). However, understanding the alignment between startup ventures and SDGs remains a complex task, hampered by limitations in manual classification methods that are time‐intensive and inconsistent. Using a dataset of over 5500 Indian startups, this study investigates how automated multi‐label machine learning and deep learning models can be optimised to classify SDGs from diverse startup information sources. We develop a multi‐modal, end‐to‐end SDG classification pipeline capable of processing website metadata, pitch decks, and pitch videos using advanced text preprocessing, classical OneVsRest classifiers, and a fine‐tuned DistilBERT model (where BERT refers to Bidirectional Encoder Representations from Transformers). This study also examines co‐occurrence patterns among SDGs and network‐based SDG clusters using frequency analysis, co‐occurrence matrices, and network diagrams. The results show that DistilBERT achieves the highest predictive performance and that Indian sustainability‐oriented startups tend to cluster into two distinct domains: an environmental cluster and a socio‐economic cluster, with only weak linkages between them. The study offers practical implications for policymakers, investors, and ecosystem enablers by providing a scalable automated tool for SDG tagging and clear visibility into thematic focus areas within the startup ecosystem. By enabling accurate SDG assessment and highlighting focus areas, the findings support more informed resource allocation and strategic interventions to accelerate progress toward global sustainability goals.