Time-Aware AutoML, ML, and DL Benchmark for Heart Disease

Description This repository provides a fully reproducible, time-budgeted benchmark framework for comparing AutoML, classical machine learning, and deep learning methods for binary heart disease detection. The benchmark integrates multiple UCI Heart Disease datasets (Cleveland, Hungarian, Long Beach VA, and Switzerland) into a unified dataset and evaluates models under strictly controlled computational budgets. The primary goal is to study the trade-off between predictive performance and computational cost in realistic, resource-constrained scenarios. The framework supports: AutoML systems (MLJAR, AutoGluon, H2O AutoML, FLAML) Classical machine learning models Deep learning architectures (CNN, RNN, GRU, LSTM, AutoEncoder) Each model family is evaluated under two runtime regimes: Lightweight regime (60-second time budget) Full regime (180-second time budget) This enables fair comparisons between fast, resource-efficient models and more computationally intensive configurations. Dataset The benchmark dataset is a merged and deterministically preprocessed version of multiple UCI Heart Disease datasets. Preprocessing includes: conversion of missing value placeholders to NaN numeric casting and schema validation binary target encoding median imputation of feature values All preprocessing steps are deterministic to guarantee reproducibility. Experimental Design The evaluation pipeline enforces strict reproducibility: fixed random seed initialization stratified train/test split inner validation split for model tuning time-budget constrained training standardized evaluation metrics Metrics include Accuracy, AUC, F1 score, Precision, Recall, LogLoss, and confusion matrices. Resource usage (runtime, memory consumption, and model size) is logged for each experiment. Reproducibility The repository is designed for complete experimental reproducibility: deterministic preprocessing pipeline fixed seeds and controlled randomness explicit dataset splits automatic figure generation saved trained models version-locked dependencies All experiments can be reproduced using the provided scripts and requirements file. Intended Use This benchmark is intended for: AutoML vs ML vs DL performance comparisons reproducibility research time-constrained model evaluation academic benchmarking studies It is particularly suitable for researchers studying efficient machine learning under computational constraints. Related Publication or Repo https://github.com/ersineneseryilmaz/heart-merged-automl-dl-ml-benchmark License This resource is released for academic and research purposes. See repository license for details.