Founder Success in Norwegian Startups: A Machine Learning Approach : A study on the use of machine learning and personality traits to predict startup performance from a pre-seed perspective

Abstract

This thesis aims to investigate founder characteristics in the Norwegian startup ecosystem and if machine learning can help venture capital firm identity successful founders at a startup’s earliest stages, when information is greatly limited. The authors collected and refined data from multiple sources, resulting in a unique dataset of 1918 tech-driven, scalable startups and 2700 unique founders. Especially outstanding in the dataset is the inclusion of personality traits estimated though the use of artificial intelligence. Four supervised machine learning models were employed to classify the founders into two created success categories, low success, and high success. The two tree-based methods, Extreme Gradient Boosting and Random Forest performed best considering the evaluation metrics, resulting in a classification accuracy of over 62%, while Logistic Regression and K-Nearest Neighbours did not follow far behind. The thesis finds significant evidence that the Number of Founders of a company and the personality trait Conscientiousness are strong predictors of success in the Norwegian startup landscape. Both of our findings showcase a positive correlation with startup performance, meaning entrepreneurs who inherits high Conscientiousness and are part of founding teams are more likely to succeed as entrepreneurs in Norway. The research has two use cases. One, to narrow the research gap on founders in Norwegian startups, and two, motivate venture capital firms in Norway to adapt and implement machine learning models to help with decision-making, despite the challenges of limited data. The authors encourage others to continue research on this area, such as investigating the validity of personality traits obtained through artificial intelligence and broadening and expanding the research to other companies in Norway and other Scandinavian countries. The thesis recognizes the potential ethical considerations that arise when collecting public data on private individuals. The weaknesses of this research are also discussed, which include the chosen data structure and biases in the data.