Code and academic paper for startup classifier. For Viktor Kozhuharov's internship in Vela Partners, with Yigit Ihlamur(@yihlamur) as an advisor.
A startup classifier that predicts whether a given startup will be successful. Uses two datasets of successful and unsuccessful companies that Vela Partners have provided.
The classifier uses logistic regression. K-nearest neighbours and random forest were also tested but performed worse. The data we use excludes any startups for which we don't have full information. The features 'name' and 'country_code' are removed because they are irrelevant ('country_code' is always USA). 'founded_year' is also excluded, as it is very hard to extrapolate patterns about specific past years to the current year. A new feature 'number_of_founders' is added. Other than being able to predict a startup's success, the model also outputs a sorted list of features by their importance.