Rating Prediction
Period: 05/2022
Project Name: Rating Prediction
Machine Learning Python os numpy random pandas tensorflow keras matplotlib plotly sklearn itertools collections
What's the Project For?
- Course Project of the "Advanced Data Mining for Risk Management and Business Intelligence" course for the degree BSc in Risk Management and Business Intelligence.
- Predict user ratings on items based on available ratings.
Project Description
We had trained and tuned 2 model architectures: Neural Collaborative Filtering (NCF) and Wide and Deep Learning (WDL). Root Mean Squared Error (RMSE) was used to evaluate our prediction performance.
Our group's final model was the NCF model with epoch = 1, embedding size = 5, and output layer using multi-layer perceptron. We achieved an RMSE of 1.0568 (< strong baseline 1.09) on the validation set.
Most Challenging Part of the Project?
This project was challenging, not because of the task, but because we had a poor assumption in the beginning of our project. In our course, we have learnt both the NCF and the WDL models, but in our project, we once ignored the NCF model because the WDL model used much more features and we thought that it would give a better model performance. We were exhausted in tuning the WDL model to meet the baseline provided by our instructor and we were far away from that even there was only 1 day left before the project deadline.
In the end, one of my groupmate realized that using the NCF model could easily pass the baseline, and we were shocked. This project had taught me that complex model does not equal to good performance, in practical tasks, we have to consider which model suits our problem, but not how complex it is. I have learnt that, sometimes, simple is the best.