L.I. Dvoyris, V.A. Gerashchenkov

The article - Evaluation of accuracy and reliability of seismic signals - detectors by using SVM method in time and frequency domains - depicts the results of theoretical researches in general capacity of learning algorithms and Support Vector Machines. On the basis of new approach to the estimation of learning algorithms capacity underlies weak axiom and the usage of it is motivated by the fact that the sample data analysis tasks must be finite. The quality of algorithms is characterized by errors - frequency in final samples that can be measured empirically by using the Monte - Carlo method. The possibility of accuracy evaluation and error estimation reliability of SVM classifier at an example of data sample classification is represented by the matrix with the observation number of 570 and six columns, 5 of them are signs of seismic signal detector and the sixth is the class sign (+1 - walking man-s signals; - 1 - different animals and natural background noise-s signals) with the presentation of data random partitions - program. The article also deals with the possibility of accuracy evaluation and classification results reliability that were obtained during the training and test samples of a given example. Probability of exact k occurrence of wrong classified objects in any training and test samples were obtained with the help of hypergeometric distribution, the Bernoulli distribution and the Poisson distribution. Computational experiments were held with the help of Mathcad sphere.