Passive Intermodulation (PIM) is a physical layer Radio Access Network (RAN) problem observed in both 4G and 5G networks. It is caused by internal physical processes such as inferior cabling or rusting, and external factors such as metallic obstacles in the radio propagation path. PIM degrades the user experience and radio resource efficiency while leading to an operation overhead for detecting and mitigating it on the operator side. Nevertheless, current solutions for PIM typically rely on costly hardware and site visit-based investigation by technicians. This work proposes a Machine Learning (ML) based PIM detection scheme for identifying PIM problems in RAN sites. Our approach relies on network KPI data already collected in the infrastructure for various purposes, including network monitoring, performance control, and maintenance. We investigate the performance of our proposed technique using empirical data collected from actual network cells.