Hazardous liquids like nitroglycerin are replacing conventional flammable explosives in modern terrorist attacks. The majority of these hazardous liquids are colorless, oily, and cannot be judged as suspicious by the naked eye. Security inspections of hazardous liquids on a large scale are urgently required to prevent terrorist activities in public areas. However, traditional liquid detection and identification techniques face issues such as cost, accuracy, and scalability, which hinder their widespread adoption. This article introduces a platform that combines radio frequency sensing based on software-defined radio technology and state-of-the-art machine learning (ML) algorithms to detect and classify suspicious and non-suspicious liquids without compromising individuals’ privacy. Specifically, fine-grained samples of orthogonal frequency division multiplexing are utilized to acquire channel state information to detect suspicious liquids (glycerin, spirit, and mustard oil) by utilizing radio signals at 900 MHz and 2.45 GHz bands. ML algorithms are employed for classification purposes based on liquids dielectric constant, and their effectiveness is evaluated based on accuracy, prediction speed, and training time. The outcomes of the performance evaluation confirm the platform’s effectiveness in accurately identifying and classifying suspicious and non-suspicious liquids with up to 98.3% accuracy with support vector machine.