Central Kurdish Linguistic Text Cryptography Dataset (CKLTCD): A Dataset for Statistical Feature-Based and AI-Driven Cryptographic Key Generation Using SHA-FFNN with AES-GCM-256 Encryption and Comparative Evaluation with PBKDF2, Argon2, and HKDF (3000 Texts)

The Central Kurdish Linguistic Text Cryptography Dataset (CKLTCD) consists of 3000 Central Kurdish texts designed for AI-driven cryptographic research. Statistical linguistic features are extracted, transformed via SHA-256, and processed using a Feedforward Neural Network (FFNN) to generate nonlinear cryptographic keys within an AES-GCM-256 framework. The dataset supports ciphertext-based evaluation with randomized salts per text and a fixed global FFNN configuration. It also enables comparative analysis with standard key derivation functions (PBKDF2, Argon2, and HKDF), where security is assessed using NIST SP800-22 tests on concatenated ciphertext streams, and key sensitivity is evaluated at the ciphertext level under one-bit key variations.

Important points:

1-All NIST SP800-22 results in this workbook were computed on concatenated ciphertext bitstreams for each method, not on generated keys.

2- Random salt values were generated for every text in all four methods.

3- A single global random FFNN (W1, b1, W2, b2) was generated and fixed for the whole experiment;

4ِِ- All parameter values are stored in sheet 03_FFNN_Random_Params. Key sensitivity was computed on the ciphertext level, using the ciphertext change after a 1-bit key modification, instead of measuring the bit change inside the key itself.