Back to selected work

PrediksiProfesi

research · 2025

PrediksiProfesi — case study

My role
Research assistant — dataset preparation, imbalanced-data experiments and model evaluation.

The problem

Profession prediction across eight classes needs training data that represents less common classes as well as the majority.

The constraints

Class imbalance makes a single aggregate accuracy score an incomplete view of the model. Synthetic examples also need evaluation before they can be trusted.

Approach and tradeoff

The research compares WGAN and class-conditional WGAN-GP approaches to synthetic data. Conditioning makes class-specific generation possible, while adding training and evaluation complexity. My work covered cleaning and analysing datasets, augmentation and resampling, and evaluating accuracy, stability and efficiency.

  1. Prepare imbalanced data
  2. Generate synthetic examples
  3. Evaluate prediction models
Research workflow overview, based on the project and CV. This is a process diagram, not a benchmark.

The result

Contributed data preparation and experiments to a doctoral profession-prediction study. The portfolio links the research code; it does not claim a numerical improvement without a published evaluation.

Inspect the evidence