Kaggle Playground
First of all, this solution is built on the idea to practise the complete workflow of machine learning prediction. Even though auto ML and blending other people's results are frequently used and legit in Kaggle, I think building a ground up solution is beneficial for gaining really solid understanding of the ML techniques.
Since my computer is not powerful enough to handle the size of data here (100K times 50) with ease, I will strive for simple and efficient way to build the classification model on Kaggle hosting server. Again, the modelling process here is for learning the ML process rather than doing the fancy stuff or building specific solutions that are not ready for generalisation to other problems.
Based on the background above, you will see a solution in favour of simple ML workflow and low computation cost, ready to be deployed for different problems. The presentation may be raw, but I will keep it to show how the result is improved gradually.
Workflow:
- Data Exploration
- Data Preprocessing
- Feature Engineering
- Feature Selection
- Model Validation And Selection
- Hyperparameter Tuning