This work presents a conservative, methodological study comparing classical machine-learning classifiers with the standard rule-based definition of Potentially Hazardous Objects (PHOs) using publicly available Near-Earth Object data from NASA CNEOS and the JPL Small-Body Database. The study emphasizes interpretability, reproducibility, and limited scope, and does not make claims regarding impact prediction or real-world hazard assessment. This preprint is shared to enable early access while awaiting archival posting.