Multi-Objective Resource-Aware AutoML for Lightweight Model Generation on Tabular Data

This repository contains the implementation and experiments for the undergraduate thesis:

"Multi-Objective Resource-Aware AutoML for Lightweight Model Generation on Tabular Data"

Prepared and submitted by:

• Dorothy C. Salva
• Jeff Lawrence C. Balbuena
• Vex Ivan Sumang

In partial fulfillment of the requirements for the degree of Bachelor of Science in Computer Science.

---

Overview

This project proposes a hybrid optimization framework for Automated Machine Learning (AutoML), designed to generate lightweight and efficient models for tabular data.

The approach integrates:

• Non-dominated Sorting Genetic Algorithm II (NSGA-II)
• Meta-learning mechanisms

to intelligently explore and optimize machine learning configurations.

---

Objectives

The system evaluates candidate AutoML configurations based on multiple competing objectives:

• Predictive Performance – Accuracy and model effectiveness
• Computational Efficiency – Resource usage and runtime performance

This enables the discovery of optimal trade-offs using multi-objective optimization.

---

Methodology

The proposed framework:

1. Uses NSGA-II to evolve candidate solutions
2. Applies meta-learning to guide the search process
3. Produces a Pareto front of optimal model configurations

---

Key Features

• Multi-objective optimization
• Lightweight model generation
• Resource-aware AutoML
• Pareto front analysis