APPLICATION OF THE BAYESIAN TRUST NETWORK AND MULTINOMIAL LOGISTIC REGRESSION TO PREDICT THE DEGREE OF CONTAMINATION OF AGRICULTURAL LAND

V.K. Kalichkin, K.Yu. Maksimovich, V.A. Shpak, R.R. Galimov, A.L. Pakul
  Download PDF
Abstract: The possibilities of using the Bayesian Network of Trust (BSD) and multinomial logistic regression (MNLR) to predict the degree of contamination of agricultural land are investigated. The probability of exceeding the economic threshold of harmfulness (ETH) with the participation of both models is calculated. Modeling of the influence of natural and anthropogenic factors using BSD was carried out, and the forecast of the excess of ETH by category was carried out using MNLR. To train the models, data from a long–term multifactorial field experience of the Kemerovo Research Institute of Agricultural Sciences - branch of the SFSCA RAS were used. Taking into account the features of the statistical sample, the main predictors of the models affecting land contamination are determined. The selected predictors were agrotechnical techniques (tillage systems, precursors) and agrometeorological resources (sums of active air temperatures, precipitation). The explained part of the variance with the Nagelkerk measure is 80.9%, which indicates high prognostic possibilities of using MNLR. The forecast results of both models coincided in 79% of cases, which indicates the achievement of high indicators of the measure of proximity of forecasts for BSD and MNLR. Both models have shown sufficiently high reliability when verified on empirical data from previous years and can be used as a tool for forecasting. The next stage of the work will be the joint use of BSD and MDR, which can contribute to strengthening the advantages of both approaches and eliminating the shortcomings of some of them.
Index terms: Bayesian networks of trust, multinomial logistic regression, forecasting, biological systems, weed vegetation, economic threshold of harmfulness.

Contacts

Russia, 659305, Altai region, Biysk,
Trofimova Street, 27, room 404B
Tel. + 7-923-162-93-27
(executive secretary -
Golykh Roman Nikolayevich)
e-mail: info@s-sibsb.ru

The certificate