School violence detection based on multi-sensor fusion and improved relief-F algorithms

School bullying is a common social problem around the world, and school violence is considered to be the most harmful form of school bullying. This paper proposes a school violence detecting method based on multi-sensor fusion and improved Relief-F algorithms. Data are gathered with two movement sensors by role playing of school violence and daily-life activities. Altogether 9 kinds of activities are recorded. Time domain features and frequency domain features are extracted and filtered by an improved Relief-F algorithm. Then the authors build a two-level classifier. The first level is a Decision Tree classifier which separates the activity of jump from the others, and the second level is a Radial Basis Function neural network which classifies the remainder 8 kinds of activities. Finally a decision layer fusion algorithm combines the recognition results of the two sensors together. The average recognition accuracy of school violence reaches 84.4%, and that of daily-life reaches 97.3%.