ML Encryption Dataset - for Classifying S-Box Generation Schemes in Embedded and IoT Cryptographic Sytems

This dataset supports the study “Real-Time Machine-Learning Framework for Classifying S-Box Generation Schemes in Embedded and IoT Cryptographic Systems.” It includes 144 S-Box samples from 12 algorithmic families, covering Dynamic, RC4-derived, and chaotic-map (Henon, Logistic) designs.

Each sample is labeled and processed into a feature vector of 40+ handcrafted attributes, capturing:

Byte distribution statistics (mean, skewness, entropy, etc.)

Transition-based features (first/second-order byte deltas)

Cyclic and positional patterns (Fourier features, sine/cosine encodings)

Information-theoretic measures (KLD, mutual information)

The dataset is structured as a CSV file with one sample per row, including the label column (algorithm class) and all feature columns. It was generated using three independent 128-bit keys per algorithm and verified for class balance.

This resource is suitable for:

Benchmarking lightweight cryptographic classification models

Studying S-Box fingerprinting in embedded systems

Teaching machine-learning pipelines for security applications

All data are anonymised and synthetically generated—no personal or hardware-identifiable content is included.