A Neural-SCHMM Approach for Delay Prediction in FANETs

In this work, we introduce the Neural Semi-Continuous Hidden Markov Model (N-SCHMM), a novel framework that integrates neural networks with Semi-Continuous Hidden Markov Models (SCHMMs) to enhance the modeling of delay distributions in Flying Ad Hoc Networks (FANETs). Our method uses a maximum-likelihood loss for unsupervised training strategy, allowing the neural component to capture complex delay dynamics. Simulation results show that N-SCHMM achieves a lower mean squared prediction error compared to the baseline SCHMM, indicating better alignment with the true delay distribution and more accurate one-step-ahead predictions. This work opens the door for improved stochastic modeling in delay-sensitive domains by combining the strengths of classical graphical models and modern neural architectures.