Scientists develop AI to predict startup success

A study in which machine learning models were trained to assess over a million businesses showed that artificial intelligence (AI) can accurately determine whether a startup will fail or be successful. The result is a tool, Venhound, that has the potential to help investors identify the next unicorn.

It is well known that around 90% of startups fail: between 10% and 22% fail in their first year, posing a significant risk for venture capitalists and other investors in start-up companies. In an effort to identify which companies are most likely to be successful, researchers developed machine learning models based on the historical performance of more than one million companies. Their results, published in KeAi’s The Journal of Finance and Data Science, show that these models can predict the outcome of a business with an accuracy of up to 90%. This means that potentially 9 out of 10 companies are correctly rated.

“This research shows how sets of nonlinear machine learning models applied to big data have enormous potential to map large feature sets to business outcomes, which is not possible with traditional linear regression models,” explains the co -author Sanjiv Das, Professor of Finance and Data Science at the Leavey School of Business at the University of Santa Clara in the United States.

The authors have developed a new set of models in which the combined contribution of the models outweighs the predictive potential of each alone. Each model classifies a business by placing it in one of the categories of success or failure with a speci fi c probability. For example, a company might have a high chance of success if the ensemble declares that it has a 75% probability of being in the IPO (listed) or “acquired by another company” category, when only 25% of his prediction would fall into the chess category.

The researchers trained the models on data from Crunchbase, a crowdsourcing platform with detailed information about many companies. They married the Crunchbase observations with patent data from the USPTO (United States Patent and Trademark Office). Given the participatory nature of Crunchbase, it was not surprising to learn that some companies’ entries lacked information. This observation inspired the authors to measure the amount of missing information for each company and use that value as an input to the model. This observation turned out to be one of the most critical characteristics in determining whether a business was going to be acquired or go bankrupt.

Lead author Greg Ross of Venhound Inc. notes that the set of models, along with new data functionality, “generates a level of accuracy, precision, and recall that surpasses other similar studies. Investors can use it to quickly assess the outlook, raising potential red flags. and make more informed decisions about the composition of their portfolios. ”

Provided by KeAi Communications