Crunchbase

Task:

Predict the status (acquired, open or closed) of a company using Crunchbase data from Kaggle. Given the strong class imbalance, our model is unlikely to surpass the baseline. We want to tune towards recall in classes 0 and 2 (in the above order). We don't mind missing out on some companies, but when we classify, we want to be classify these classes correctly.

Key methods/challenges:

• Classification with imbalanced classes - under sampling, SMOTE, pipelines
• Creative-ish feature engineering

Results:

• ~60% recall in classes 0 and 2.
• Will need more work to optimise!