Analyzing Remote Work Productivity
An R-based classification study identifying what actually drives remote work productivity, comparing five algorithms across five feature-selection methods.
The Problem
Remote-work survey datasets are wide and noisy — most attributes don't matter, and picking a single classifier without testing feature-selection strategy against it risks tuning the wrong thing.
Why I Built It
A feature-selection-focused project to get precise about the difference between an attribute that correlates with productivity and one that actually predicts it.
What I Built
Ran an R-based analysis using five classification algorithms (Rpart, J48, KNN, Gradient Boosting, and Naive Bayes) against a remote-work productivity dataset, applying five attribute-selection methods (Chi-Square, RFE, Gini Index, Information Gain, and ReliefF) to compare which feature sets actually improved classification accuracy.
Technical Challenges
Cross-referencing five classifiers against five feature-selection methods means 25 combinations to evaluate fairly — the real work was building a consistent evaluation harness so accuracy differences reflected the algorithm/feature-set pairing, not inconsistent preprocessing between runs.