Crop weed infestation forecasting using data mining methods

Abstract

The studies were carried out to develop logical rules for crop weed infestation forecasting using data mining methods within limited sampling conditions. The classical probabilistic-statistical methods and decision tree method were used. Selected methods were chosen considering the distribution of the input data, the diversity of its factors and attributes, the relationship structure peculiarities. The analysis was performed on the extensive field experiments data of the Kemerovo Scientific Research Institute of Agriculture – a branch of SFSCA RAS on weed infestation of agricultural soils by cereal weeds for 2013–2019. The qualitative factors (tillage system and first crop) and meteorological features (average ten-day air temperatures and amount of precipitation) determining the indicators of crop weed infestation were outlined. The statistical significance and contribution rate of each factor were evaluated. The decision tree based on the set of logical rules was built, which makes it possible to forecast the weediness index. The coefficient of determination of the model was 0.68, which is a sufficiently successful result for the forecast of the biological nature object. The results obtained (information about relations between indicators and factors, a set of logical rules) can be used in the design of knowledge-based decision-making support systems in crop production.