Application of Predictive Techniques for Startup Survival: The Ecuadorian Case

Failure is common among startups, highlighting the need for more effective strategies to improve their chances of survival. This study examines three advanced survival analysis methods: Random Survival Forest (RSF), the Cox proportional hazards model, and survival analysis using Gradient Boosting. The goal is to model and predict the duration of business activities and the factors influencing the closure of startups in Ecuador, comparing the accuracy of these methods to determine the most robust. The data comes from the Statistical Business Register of the National Institute of Statistics and Censuses, spanning from 2012 to 2020, and was preprocessed to resolve inconsistencies and missing values using imputation techniques. Among the findings, it is noted that RSF, using multiple decision trees, showed limitations in identifying nonlinear relationships in this data. The Cox model, known for its clear interpretation of covariates, indicated that variables such as total sales and the number of employees were not significant, while geographic location showed high variability and instability in predictions. The Gradient Boosting Survival Analysis method, iteratively optimized, demonstrated the best predictive capability, accurately capturing the complex and nonlinear relation-ships between variables and the time-to-event. This approach suggests that boosting models are particularly effective for the survival analysis of startups, offering a valuable tool for entrepreneurs, investors, and policy-makers seeking to optimize strategies to enhance business survival.