Using big data in startup selection

: exploring machine learning as a tool to predict successful startups in the age of social media

Abstract

This research aims to further explore the possibilities in the usage of Machine Learning within the Venture Capital industry. Building on previous research the goal of this paper is to determine whether social media analyses can improve the accuracy of Machine Learning models to predict startup outcomes and valuations for startup companies. The research is built on the following models: Multilayer Perceptron, XGBoost, RandomForest, Naive Bayes, and Voting Regressor. The data used in this research comes from Crunchbase, USPTO, and Twitter. The models in this research achieved an adjusted R2 of 0.5281 for value prediction, whichshows that exit value is explainable to a large extent by using publicly available qualitative and quantitative data. Outcome prediction had precision for IPO between 0.1447 to 0.4193and F1-scores between 0.2360 to 0.4449 for models built from Series A to Series C funding rounds. The results of this research show that Venture Capital firms investing from Series A to Series C would be able to outperform the market in terms of returns by implementing Machine Learning in their investment decision-making process. To further improve these results extracting further social media data is a beneficial future resource. Compared to previous models this research built models for 3 specific early funding rounds and can outperform the markets with data available for VCs at these points in time.

Date of Award	19 Oct 2022
Original language	English
Awarding Institution	Universidade Católica Portuguesa
Supervisor	Alessandra Luzzi (Supervisor)