Abstrak
Assessment of the performance of the teaching assistant as a form of real behavior displayed by each assistant as a work achievement is carried out every year. The evaluation parameters of lecturer assistant performance are taken from the UCI Machine Repository where these parameters are English speakers, course instructors, courses, summer or regular semesters and class sizes. The evaluation process requires a model as the best feature selection, in this case we propose a co-evolutionary particle swarm optimization method to improve the accuracy of the mechine vector supprot method. Testing the dataset using the Rapid Miner software of various validation criteria starting from the accuracy test, precission test, recall test and then presented in the AUC curve. The results of the development of co-evolution based mechine vector-particle swarm optimization methods provide good classification and increase the validation value so that it can be used as a periodic control in evaluating the performance of the teaching assistant.