P-ML: An End-to-End AutoML Framework for Deploying Classical Machine Learning Models on Resource-Constrained Devices

P-ML is an end-to-end AutoML framework for deploying classical machine learning models on memory-constrained microcontrollers. The framework automates the complete workflow, including data splitting, model selection, hyperparameter optimization, and generation of optimized Arduino-compatible C++ libraries. P-ML integrates Optuna-based hyperparameter tuning with stratified data splitting methods such as SPXY and K-Fold cross-validation to ensure robust and reliable model selection. The generated libraries are compact, efficient, and directly deployable on Arduino Uno, Nano, and ESP32 platforms using the Arduino IDE. Experimental results demonstrate that P-ML enables accurate sensor data classification, achieving over 90% accuracy while maintaining a small memory footprint suitable for embedded IoT applications.