In order to solve the problem that the basic particle swarm (PSO) to optimize the parameter of SVM is easy to fall into local optimum, this paper proposes a modified classifier that uses the improved particle swarm optimization (IPSO) to optimize the parameter of SVM (IPSOSVM) by introducing the new dynamic inertia weight, global neighborhood search, shrinkage factor and mutation operator of genetic algorithm. The classification result is tested by the common datasets named Iris, Wine and seeds from UCI machine learning repository, the result shows that IPSOSVM classifier is better than GSSVM, AFSASVM, GASVM and PSOSVM classifier in terms of classification accuracy and classification time. The better convergence ability and speed of the IPSOSVM classifier are verified by fault diagnosis of SallenKey bandpass filter, fouropamp biquad highpass filter and nonlinear rectifier circuit.