Skill-Based Salary Prediction for Data Roles in the Philippines
Undergraduate thesis, Mapúa University · Project Lead, Programmer
- Collected Philippine data job postings from Indeed with Apify over four months (Nov 2024 – Mar 2025), then deduplicated, filtered salaries and normalized titles down to 1,057 salaried postings.
- Used GPT-4o-mini to extract and standardize soft and technical skills from unstructured job descriptions, yielding 720 distinct skills.
- Grouped 43 job titles into 6 role families with Sentence-BERT embeddings and K-Means, testing k with silhouette scores.
- Compared Linear Regression, Random Forest, LightGBM and XGBoost for predicting min and max salary, with grid search and 5-fold cross-validation. XGBoost did best on the held-out test set.
- Converted model feature importance into estimated peso values per skill to show which skills go with higher-paying roles.
// analysis showcase
- Salaried postings
- 1,057
- Collection window
- 4 months
- Job titles → role families
- 43 → 6
- Distinct skills extracted
- 720


AscendyX
Rizal Commercial Banking Corporation (RCBC)
Mapúa Radio Cardinal