Hacker Earth Machine Learning Competition
- Merge user's submission log with user and problem details
- Created new features like num_attempts at a problem, only took 200,000 examples from around 400,000 available.
- Created some features related to user's problem solving capability, how hard a problem is and used rest of the features specified in the examples
- One hot encoding of user's skills
- Label Encoding categorical Features
- Trained Logistic Regression Model, Random Forest Classifier, Extra Trees Classifier, SGD and XGBoost Model
- But final model was an Extreme Gradient Boosting Model with parameters that produced the best accuracy on the cv set