Refined Proposal

We will use LSTMs to analyze sequential patterns in startup development, including funding rounds, hiring changes, and other time-dependent signals. This approach differs from the original research paper, which did not model temporal data.

For datasets, we will combine multiple sources such as Kaggle startup datasets, Crunchbase information, and SEC filings. Using several sources will give us a larger and more diverse dataset than the original study and will allow us to capture more detailed startup trajectories.

Instead of predicting a simple success or failure outcome, we will classify startups based on their eventual path. Examples include acquisition, unicorn status, shutdown, IPO, or continuing operation. This multi-class setup reflects the more realistic range of outcomes that startups experience.

We will be using the paper “Deep Learning to Predict Startup Business Success” as our guideline and baseline for comparison.

Finally, we will compare our LSTM-based model with the architecture and results presented in that paper to evaluate whether sequential modeling, expanded datasets, and more nuanced labels lead to improved performance.

Built With

• python