Acceleration Signal Categorization Using Support Vector Machines

Large sensor installations are becoming prominent as the cost of sensors drop and new methods are developed for structural health monitoring, fall detection, building occupancy, etc. Large amounts of data could be quickly captured, especially for measurements of high sampling rate such as acceleration signals. Methods to quickly triage records for further analysis can be used to drastically reduce the amount of data to be process. This paper studies the use of Support Vector Machines to classify floor vibration signals to determine signals of interest. Four kernels and three signal metrics were explored in this research using a human activity dataset containing over 500,000 acceleration records. Results show that the Radial Basis Function using a Dispersion Ratio metric can be used to identify signals of interest effectively.