03 / Automated machine learning
MetaRecom.
A Flask-based AutoML platform that profiles tabular datasets and recommends suitable classification or regression algorithms using meta-learning.
01 / OVERVIEW
The context.
MetaRecom profiles a tabular dataset, derives descriptive meta-features and uses similarity-based meta-learning to recommend an algorithm family. It also provides configurable hyperparameter optimization and hardware-aware execution modes.
02 / CONTRIBUTIONS
What I worked on.
- Built a 32-feature dataset-profiling pipeline covering statistical distributions, missing values, correlations, PCA-based descriptors and outlier indicators.
- Implemented a KNN-based meta-learning recommender with distance-weighted aggregation.
- Integrated RandomizedSearchCV for automated hyperparameter optimization.
- Added configurable GPU/CPU execution modes with XGBoost CUDA capability detection and CPU fallback handling.
03 / ENGINEERING NOTES
How it comes together.
01The recommender uses dataset characteristics as meta-features to compare a new dataset with previously observed dataset profiles.
02The platform combines a recommendation stage with a conventional hyperparameter-search stage instead of treating model selection as a single fixed operation.
04 / TECHNOLOGY