Classifying high-potential startups for strategic partnerships using machine learning – The case of german digital startups

Traditional methods for identifying strategic partners often rely on human judgment and informal networks, which would likely cause poor, strongly-biased choices, and missed opportunities. In this study, we aimed to support partner selection through a machine learning approach using publicly available startup data. First, we addressed class imbalance using SMOTE and generated negative labels through heuristic clustering. A Logistic Regression model was then trained to classify high-potential startups based on features such as technological capabilities, funding phase, innovation hub association, founders' structure, and geographical proximity. As for the results, the model achieved solid performance with 88.2% accuracy, 80% precision, and an F1-score of 0.727. Key patterns showed that technology prowess had the most prominent impact on defining a 'high-potential' startup partner, closely followed by their funding stage and close geographical distance to the focal company. This study provides a replicable, data-driven framework to implement into the early partner screening stage for companies, especially SMEs, along with insights on selection criteria for startups.