Search

Search Results

• Identifying maize yield drivers using statistical analysis and multilayer perceptron modelling

  Péter Fejér

  Adrienn Széles

  Péter Zagyi

  Éva Horváth

  Views:

  19

  Maize yield formation is jointly determined by crop-year conditions, water availability, nutrient management, and their interactions, which may include complex nonlinear relationships. This study aimed to identify the principal agronomic and physiological variables associated with maize yield using classical statistical analyses complemented by multilayer perceptron modelling. Field data collected in Debrecen, Hungary, during 2024–2025 included crop year, irrigation, fertilizer treatment, phenological stage, SPAD chlorophyll readings, and grain yield. Pearson correlation and linear regression quantified individual relationships, while four MLP scenarios evaluated combined predictive effects. Fertilizer was positively associated with yield in both years and irrigation regimes, with the strongest relationship under irrigation in 2025 (r = 0.770; R² = 0.594). The SPAD–yield relationship generally strengthened during crop development, reaching its maximum at R3 under irrigation in 2025 (r = 0.916; R² = 0.839). The best-performing MLP scenario included year, fertilizer, and irrigation, confirming their central role in yield prediction and formation.

• Predicting maize yield with a multilayer perceptron (MLP) model using multivariate field data

  Péter Fejér

  Adrienn Széles

  Péter Ragán

  Csaba Juhász

  Éva Horváth

  Tamás Rátonyi

  Views:

  214

  This study presents the findings of a multi-year maize field trial conducted on experimental plots between 2017 and 2019, focusing on the application of machine learning techniques to enhance yield prediction accuracy. A multilayer perceptron (MLP) neural network was employed to model the effects of agronomic treatments, environmental variation, and compositional traits. Six distinct modeling scenarios were developed to explore different combinations of input variables, with the grain yield of maize serving as the sole output parameter. These scenarios range from treatment-only models to those incorporating detailed quality and compositional data. The primary objective was to evaluate how well MLP models can capture the complex, nonlinear relationships influencing yield under varying conditions. The findings provide valuable insight into the role of machine learning in supporting decision-making for sustainable crop production, especially under diverse technological and environmental settings. The approach demonstrated here offers a foundation for more adaptable, data-driven strategies in agronomic optimization.