Modeling and classifying neuronal activity with a fusion of mathematical and machine learning techniques

Predicting neuron spike patterns is crucial because spikes are the brain's fundamental language, revealing how information is encoded and transmitted. Such prediction also supports disease diagnosis, brain-machine interfaces, and the control of robotic arms, wheelchairs, and neuromorphic AI design. Yet, simulation techniques often suffer from limited biological realism, numerical instability, and poor generalization across diverse neuronal activity types. These limitations are further compounded by the scarcity of high-quality, labeled datasets that capture the full spectrum of neuronal dynamics, restricting the training and evaluation of machine learning models. To address these challenges, we proposed SpikeNet, a hybrid framework that integrates the Izhikevich neuron model with the Runge-Kutta fourth-order (RK4) algorithm to generate synthetic voltage signals that are both biologically plausible and computationally precise. These signals are then used to train a Bidirectional Long Short-Term Memory (Bi-LSTM) network, which effectively captures long-range temporal dependencies in spike trains. SpikeNet combines accurate simulations with advanced sequence modeling to improve spike pattern classification, providing a scalable solution for reliable data generation and prediction. The proposed model was evaluated on multiple datasets, including single-spike data, multi-label spike data, and the Allen dataset, and demonstrated strong performance across all evaluation metrics.