Machine learning and bifurcation analysis in a discrete predator-prey model with neem-induced mortality.

This study develops a discrete-time predator-prey model for guava pest management using the piecewise constant argument (PCA) scheme. The model incorporates logistic prey growth, neem-induced mortality, and predator crowding. Analytical and numerical results establish conditions for flip and Neimark-Sacker bifurcations, supported by bifurcation diagrams, Lyapunov exponents. Ecologically, small neem-induced mortality (d) destabilizes prey-predator coexistence, whereas larger d restores stability. The intervention frequency [Formula: see text] further shapes dynamics, with moderate values maintaining stability and large values inducing oscillations. As a proof-of-concept, machine learning (random forest and decision tree classifiers) was explored to efficiently approximate the analytically derived stability regions. Both classifiers successfully replicated the stability map, with Random Forest providing smoother boundaries and higher accuracy, demonstrating the potential of ML as a computational surrogate for more complex models. Parameter importance analysis revealed that prey dynamics are mainly driven by prey-related parameters (r, a, b), while predator persistence is strongly influenced by conversion efficiency (c) and natural mortality (s). These findings highlight that balanced neem application, appropriate timing of interventions, and conservation of natural enemies are key for sustainable guava pest control.