Amazon Machine Learning Challenge is a competition for all engineering students, researchers to showcase their talent and find creative and innovative solutions to real world problems. The dataset provided by Amazon had 3 million rows and the challenge was to use that data and come up with an efficient model to classify products into different nodes. This is our solution to Amazon ML Challenge 2021 which fetched us a rank 30 briefly on the first submission itself.
Dataset was provided by Amazon and was a csv file containing more than 3 million rows of raw data
- Cleaning the data by removing all links, patterns, emojis etc.
- Creating new features using 'Title', 'Bullet Points'.
- Shuffling the data to get better sampling.
- Taking subset of each class for training
We started with classical ML algorithms which served us well and then moved on to other algorithms and models.
Classical models we tried:
- KNN
- Random Forest
- SVM
- Naive Bayes
Out of these KNN gave the best output
Deep learning based models we tried:-
- Tried a Conv1d based architecture we created ourselves.
- Deep Neural Network
- BERT transformer.