The improved particle swarm optimization (IPSO) optimized hyper-sphere support vector machine (HSSVM) can be used for abnormal detection of keyboard in this paper. Firstly, the development of the hook (hook) procedure in the Windows operating system is used to collect the required key time series as a training set and test set through the system messages WM_KEYDOWN and WM_KEYUP capture keyboard keystroke messages. Then, the HSSVM model is used to carry out sample training and finally transformed into a quadratic programming problem. The IPSO is used to optimize the penalty factor and kernel parameters of HSSVM model. Finally, the test set is used to verify the accuracy of the model detection and is compared with the results before optimization. The test results show that the IPSO-HSSVM model is effective for the detection of the keyboard and the accuracy rate is over 90%, which is better than that of the HSSVM before optimization. However, it is necessary to further improve the quality and quantity of the training samples in order to obtain higher detection accuracy.