Seismic sensors are widely used to monitor human activities, such as pedestrian motion and detection of intruders in a secure region. This paper presents a symbolic dynamics-based method of data-driven pattern classification by extracting the embedded information from noise-contaminated sensor time series. In the proposed method, the wavelet transforms of sensor data are partitioned to construct symbol sequences. Subsequently, the relevant information is extracted via construction of probabilistic finite state automata (PFSA) from symbol sequences. The patterns are derived from individual PFSA and are subsequently classified to make decisions on target classification. The proposed method has been validated on field data from seismic sensors to monitor infiltration of humans, light vehicles, and animals. The results of pattern classification demonstrate low false-alarm/missed-detection rate in target detection and high rate of correct target classification.