Predicting Pollen Germination and Tube Elongation Responses to Different Plant Growth Regulators in Kiwifruit (Actinidia deliciosa L.) Using Random Forest Regression

Abstract

This study aimed to train a Random Forest regression model using pollen germination rate and pollen tube length data obtained after 3 h of in vitro germination at 0.005, 0.05, and 0.5 mM concentrations of 24-epibrassinolide, methyl jasmonate, spermidine, spermine, and putrescine, and to evaluate the model’s accuracy in predicting responses at 0.025, 0.25, and 2.5 mM concentrations. Experimental data were compared with Random Forest Regression model predictions, and model performance was assessed using Absolute Error and Root Mean Square Error. Prediction accuracy was classified as good, moderate, or low based on Absolute Error thresholds applied to both pollen germination and pollen tube length (0-6, 6-15, ≥15), and Root Mean Square Error thresholds defined separately for pollen germination (0-10, 10-20, ≥20) and pollen tube length (0-20, 20-40, ≥40). Results indicated that the Random Forest Regression model provided reliable predictions at low and moderate plant growth regülatör concentrations, with 24-epibrassinolide and putrescine treatments aligning closely with experimental data. However, for methyl jasmonate, spermidine, and spermine at higher concentrations, the model exhibited overestimations, particularly in predicting pollen germination rates at inhibitory doses. The study highlights the potential of machine learning approaches in pollen biology research and demonstrates the necessity of optimizing model parameters for high-dose predictions. These findings contribute to the integration of data-driven decision-making in artificial pollination and plant growth regulators treatment strategies.

Code and data availability

公開論文であることは確認できましたが、現在の公式API・許可済み取得経路では本文を自動取得できませんでした。

No evidence-backed public reproduction asset is currently recorded.