Predict the default

Companies like machines can sometimes fail aka default on their obligations. We measure a company’s failure as it’s the ability to service its debts.

The challenges require you to build a machine learning model to predict this probability of default for a given set of instances. An instance is basically a snapshot of the company at a given point of time

Data Precision

We provide you with 33 anonymous features to help you build this model. The goal will be to beat the benchmark given below. Maximise recall where precision is at-least 0.2

Data Preview

The data set for training is available for download here.

  • To test for your performance of the model, split the training into train and validation or cross-validation approach, whichever you prefer.
  • Make sure to split in such a way that the IDs mentioned are exclusive to either train or validation. E.g. All instances of ID 40813 belong either to train or validation.
  • Check for performance across your validation set for recall-precision values.


If you better the above-mentioned benchmark, make your predictions against this Test Set available for download here. Then send the output which has 62,601 instances in the same order to


We will get back to you if you are able to generalise your performance over this data set as well.

  • To incentive participation, we have a prize simple price structure of $500, $300 and $200 dollars for the best 3 entries beating the benchmark.
  • We might still reward good performance but not benchmark beating if the approach is sufficiently orthogonal to the status quo but is at the complete discretion of the organizers.
  • The contest runs till the end of March 31st 2020.
  • To be able to claim the prizes, you will transfer any kind of ownership or royalty rights over the solution/approach to ASQI.