Wage Prediction: A Regression Competition
This DS project is a result of the 7th FLAI Machine Learning Competition. The aim of the contest was to build a model that predicts wage with the smallest mean absolute error.
This model won 1st place 🏆 in the Competition.
For a better understanding, this was divided in 3 parts: Data Analysis, Machine Learning Algorithm and Deploy. The last one was not part of the competition but a way to practise the last DS step.
Note: Although the code cells are in english, the explanation and result interpretations (markdown cells) are in portuguese.
- Exploratory Data Analysis: EDA
At this stage, the goal was to understand the datasets.
- ML: Algorithm
Following the EDA, a pre-processing was carried out. For this, a randomized search over all the possible transformations (that I could think of at the time) was implemented.
The last 5 submissions (of a total of 10) used the transformations chosen by this method.
Thereafter, randomized searches were carried out for a hyperparametrization.
Finally, a voting regressor using the best models was the one employed to predict the wage.
- Streamlit model deploy: Deploy
In this final step, joblib was used to save the model. And then, streamlit for the interface.
Note: Despite the fact that the EDA found some duplicated data, the deploy did not take this into account as well as the 'clustering'