Predicting start-up success using a machine learning

The limited availability of historical financial data and performance indicators makes it considerably difficult to reliably assess the success potential of start-ups. This study aims to predict the success of start-ups operating in Turkey by employing machine learning methods. Within the scope of the research, both quantitative and qualitative variables related to start-ups were analyzed through a data-driven approach to identify the factors influencing their success. In the data collection process, data from Crunchbase and Startups.watch platforms were systematically gathered for start-ups founded in Turkey between 2018 and 2022. Various data preprocessing and feature engineering techniques were applied to construct a final dataset ready for modeling. During the implementation phase, predictive models were developed using seven different machine learning classification algorithms: Random Forest, AdaBoost, Gradient Boosting, Voting Ensemble, LightGBM, XGBoost, and RUSBoost. Among these models, the Random Forest algorithm yielded the highest performance, achieving an accuracy rate of 85.09%. The findings of the study revealed that the number of Twitter and LinkedIn followers, as well as website rankings, are among the most significant determinants of start-up success. These factors are followed by the number of employees, number of investors, and company age. The study underscores that performance in the domain of digital visibility is as critical as financial data for start-up success. Additionally, variables such as the number of investors and the age of the start-up are also identified as key influencing factors. Conducted within the context of the Turkish start-up ecosystem, this study provides a unique contribution to the literature.