The internet of things (IoT) connectivity scenarios are becoming increasingly diverse as well as the types of connected terminals. IoT services are mainly carried on telecom operators’ cellular networks, which enable applications in various scenarios while also causing significant issues in service and network perception for operators. This paper proposes an automatic detection method for IoT service perception anomalies based on machine learning. It utilizes the variational autoencoder (VAE) algorithm to encode and decode IoT service perception data, and performs anomaly detection through the different fluctuation amplitudes formed during the encoding and decoding processes. After practical verification, this technology can accurately detect IoT service perception anomalies after establishing an appropriate reconstruction error threshold. It can help operators discover IoT industry perception issues more quickly and accurately, providing a referential solution for IoT industry perception detection issues.