This project aims to predict the performance of Y Combinator startups using survival analysis techniques in machine learning and statistics. Our goal is to understand the key factors that contribute to the success of startups in the Y Combinator ecosystem and to explore what criteria Y Combinator looks for when evaluating potential startups for acceptance.
The inspiration for this project came when a friend of ours was rejected from Y Combinator. We decided to leverage our skills in statistics and machine learning to investigate the underlying patterns and criteria that could help aspiring founders like us improve their chances of success.
 Alisha Srivastava |
 Siddanth Arvind |
We are utilizing a Kaggle dataset on Y Combinator startups, augmented with additional features we built using external data from scraping & APIs. This includes funding data, founder characteristics, and program-specific metrics such as the YC batch size and time to acquisition. https://www.kaggle.com/datasets/sashakorovkina/ycombinator-all-funded-companies-dataset/data?select=founders.csv
We are currently integrating exploratory data analysis (EDA) with survival-specific methods, such as Kaplan-Meier survival curves. Feature engineering is ongoing, focusing on creating meaningful predictors for the survival models.
Primary Survival
- Definition: A startup is considered to have survived if it is acquired or remains operational (not defunct) after completing the YC program.
- Event Indicator: Binary (1 if acquired, 0 otherwise). Secondary Survival
- Definition: A startup is considered to have survived post-YC if it raises Series A funding or higher after completing the YC program.
- Event Indicator: Binary (1 if Series A funding is raised, 0 otherwise).
YC-Specific Feature: Series A funding and operational status is considered only after the YC end date (start date + 3 months).
This project combines statistical modeling and machine learning to analyze survival outcomes: Statistical Modeling: Cox Proportional Hazards Model: For identifying statistically significant predictors of survival. Machine Learning: Random Survival Forest: For capturing non-linear relationships and complex interactions between variables.
After completing the EDA and regression modeling, we plan to:
Model Development:
- Fit and evaluate Cox Proportional Hazards and Random Survival Forest models.
- Compare performance metrics such as C-index and log-rank test results.
Actionable Insights:
- Identify the most influential factors affecting startup survival.
- Provide recommendations for aspiring YC founders based on the findings.
Portfolio Showcase:
- Incorporate visualizations and dashboards into the GitHub repository to enhance usability and presentation.
To run this project, clone the repository and ensure you have the necessary libraries installed. You can install the required packages using the following command:
pip install -r requirements.txt