jquintanac / proj-ml

Class project: a machine learning predictions about salary in data jobs.

Machine learning - Kaggle competition

The goal of the competition is to obtain the lowest RMSE from machine learning predictions of a data jobs.

Kaggle

Competition bases

• Predict the salary for data Jobs with machine learning
• Data:
  • Salaries_data.csv (Working data)
  • Testeo.csv (Predicting data)
  • Muestra.csv (Example for Kaggle)
• Data will be uploaded to Kaggle and RMSE will be calculated as a score.
• Lowest score wins the competition

ETL

• Salary and salary_currancy: both were eliminated from the table (salary_in_usd, the column-to-predict, proceed from both).

• Company location: Grouped by 'US' or 'Not US'. Too many values to get a good correlation value and to work with.

• Employee residence: Grouped by 'US' or 'Not US'. Too many values to get a good correlation value and to work with.

• Employment type: Grouped by 'FT' or 'Not FT'. Too many values to get a good correlation value and to work with.

• Job tittle: restricted to 'Data Scientist’, 'Data Analyst’, 'Machine Learning Engineer’, 'Research Scientist’ and ‘Other’

• Categorical to numeric. With 'get_dummies', I transformed categorical columns to numeric colums to work with them.

Correlation values were improved with these changes.

Predictions

RMSE values were obtained with its own training:

• Random forest: 57288.29736536037
• ExtraTreeRegressor: 54998.40405193934
• LinearRegression: 55869.57667978832
• Lasso: 55866.50186298033
• Ridge: 55786.59417475289
• ElasticNet: 60719.37461777011
• SVR: 74151.47731646513
• LGBMRegressor: 56461.887229577005
• XGBR: 60513.10436434398
• CTR: 59449.57853642741
• GBR: 56630.53859375318

Final result

🥉 Third ranking position from my class!