Application of Machine Learning Algorithms in
Disaster Risk Forecasting
Mir Ramin Yunusov1* , Yalchin Zeynalov2 ,
Ibrahim Mazanov3 , Gafur Huseynov3
Abstract. The escalating cascade of natural and anthropogenic hazards, exacerbated by global climate change and intricate industrial infrastructures, requires a fundamental reassessment of mathematical-analytical models in emergency response. This research mainly aims to comparatively evaluate of linear and non-linear mathematical equations that depict the spatio-temporal propagation dynamics of disaster risks. The analytical framework applies dynamic systems theory, ordinary and partial differential equations, bifurcation analysis and the mathematical modeling of cascade effects. Analytical practicalities and scenario simulations indicate that in cases restricted occurrences — including floods or minor landslides — linear autoregressive models and simple differential equations deliver reasonably precise and quick risk assessments. Conversely, in scenarios of swiftly developing, complicated disasters involving critical infrastructure failures, linear models demonstrate significant limitations. Modeling such asymmetric risks, disaster "tipping points," and exponential spatio-temporal propagation demands the application of reaction-diffusion equations and coupled non-linear systems. The results scientifically prove the boundaries of linear models in capturing real-world dynamics and validate the clear superiority of non-linear approaches in complex disaster scenarios.
Keywords: machine learning, cascade effects, non-linear dynamics, mathematical modeling, disaster risk, differential equations, tipping points, crisis management