FogMLS: A Labeled Dataset for Machine Learning-Based Resource Scheduling in Heterogeneous IoT-Fog-Cloud Systems.

FogMLS is a collection of labeled datasets supporting supervised machine learning research in fog computing resource scheduling. Three datasets are provided, covering low (3x3), medium (3x6), and high (3x9) IoT device density configurations, each generated under random seed 42 using the FogMLS simulator with a genetic algorithm scheduler and greedy initialisation. Dataset structure per file:Rows: 5,000 (one per scheduling slot)Feature columns: 600 (200 task positions x 3 features each)Label columns: 200 (one resource assignment per task position)Total columns: 800Missing values: 0File format: CSV Features per task position:Task type code (0, 1, 2, 3)Instruction count in simulation unitsData size in simulation units Label values:0: Padding (unused task position)1: Cloud server (CPU: 48,000 SU, latency: 50 SU)2: Fog-1 (CPU: 40 SU, RAM: 2,000 MB, BW: 2 SU)3: Fog-2 (CPU: 80 SU, RAM: 4,000 MB, BW: 4 SU)4: Fog-3 (CPU: 160 SU, RAM: 8,000 MB, BW: 8 SU) The primary high load dataset (3x9) contains 404,919 active task-resource assignments across 5,000 slots. A multi-output deep learning pipeline validated the labels, achieving 93.47% overall accuracy with F1 scores above 0.88 for all four scheduling classes. Label consistency was confirmed across three independent random seeds (42, 123, 456). This dataset accompanies a companion paper currently under review at Scientific Data (Nature Portfolio).