The rapid development of the Internet of Things has led to the widespread use of sensors in everyday life. Large amounts of data through sensing devices are collected. The data quantity is massive, but most of the data are repetitive and noisy. When traditional classification algorithms are used for classifying sensor data, the performance of the model is often poor because the classification granularity is too small. In order to better data mine the knowledge from the Internet of Things data which is a kind of big data, a new classification model based on subspace probability detection is proposed. This model can be well integrated with traditional data mining algorithms, and the performance on sensor data mining is greatly improved.