TYPING OF HOUSEHOLD PLOTS BY THE METHOD OF NEURAL NETWORK CLUSTER ANALYSIS

DOI 10.32651/236-97

Issue № 6, 2023, article № 14, pages 97-107

Section: Problems of agroeconomic researches

Language: Russian

Original language title: ТИПИЗАЦИЯ ЛИЧНЫХ ПОДСОБНЫХ ХОЗЯЙСТВ МЕТОДОМ НЕЙРОСЕТЕВОГО КЛАСТЕРНОГО АНАЛИЗА

Keywords: SELF-ORGANIZING MAPS (SOM), CLUSTER ANALYSIS, NEURAL NETWORKS, TYPIFICATION, HOUSEHOLD PLOTS

Abstract: The article provides a methodology for typing household plots using a special model of artificial neural network – self-organizing map (SOM). To reduce the number of features and visualize the results of typing, three multivariate means are calculated based on indicators of the state of infrastructure in rural areas and logistical support, indicators of the size of crop areas and the presence of livestock. In order to stabilize the dispersions of feature values and ensure better formation of clusters as typical groups of households, sharply different feature values are determined and excluded based on the interquartile range method. Approbation of the methodology was carried out according to the microdata of household plots of the Astrakhan region of the All-Russian Agricultural Census of 2016 (VSHP-2016): eight qualitatively different and internally homogeneous groups of households in terms of size and production direction were identified. The statistical significance of the influence of the used multivariate means on the formation of clusters has been established by analysis of variance. The article compares the results of applying the SOM model and the k-means method, formulates a number of advantages and disadvantages of each of the approaches. The use of the developed typology will allow developing and implementing a differentiated state policy aimed at the inclusive development of agriculture, targeted support for household plots.

Authors: Ukolova Anna Vladimirovna, Bykov Denis Vitalevich