DOI: 10.17587/prin.9.82-90

The Effect of Dimension Reducing on Classification Results of Leaves of Various Plant Species

V. A. Kharakhinov, e-mail: tes4obse@mail.ru, S. S. Sosinskaya, e-mail: sosinskaya@mail.ru, National Research Irkutsk State Technical University, Irkutsk, 664074, Russian Federation

Corresponding author: Sosinskaya Sophia S., Professor, National Research Irkutsk State Technical University, Irkutsk, 664074, Russian Federation, E-mail: sosinskaya@mail.ru

The article considers the process of classification of different plant species, described by a set of numeric properties, based on multilayer perceptron. It was proposed to apply the factor analysis to reducing dimension of the original data set. Two well-known criteria for determining the number of factors was used. Such as Kaiser criteria and variance explained criteria. The neural networks were trained on native data set and on factor scores to compare time expenditures of training process. The quality of classifications for both cases is displayed in tables. The graph displays time expenditure decreasing for network training when factor scores were used. Also the graphs display relation of classification error rates from a number of factors. These graphs allow one to conclude which network, with proper selection of the number of factors, provide an efficient way to reduce the time expenditure of training process, at the same time allow achieving a high quality of classification.

Keywords: classification, feedforward network, machine learning, factor analysis

pp. 82–90

For citation:
Kharakhinov V. A., Sosinskaya S. S. The Effect of Dimension Reducing on Classification Results of Leaves of Various Plunt Species, Programmnaya Ingeneria, 2018, vol. 9, no. 2, pp. 82—90.