This work aims at analysing different aspects of the nonlinear dynamic behaviour of the ME heterostructure which is composed of magnetostrictive Cr-FeCoSiB and piezoelectric AlN thin films, having a natural frequency of 7.4kHz, by modelling its timeseries readout with several machine learning algorithms. The excitation conditions were varied in terms excitation frequency and the amplitude of bias field while magnetic excitation was performed along the easy axis of the cantilever. The sensor was operated under linear conditions and the nonlinearity originating from the amplification (e.g., power, charge) systems were focused on modelling. The optimized model-equations were further analyzed indicating a linear damped oscillator for the harmonic vibration case and self-excited nonlinear damped oscillator for the super harmonic vibration (i.e., 1/3 of the resonance frequency) case.The discovered equations suggested the existence of only the trivial fixed points, i.e., (0, 0) point in the phase space of the systems, for which the existence and stability of the trivial and non-trivial steady state amplitudes were examined via local bifurcation analysis by considering the magnitude and sign of the as-discovered linear and nonlinear damping coefficients and were verified with the phase portraits of the individual dynamic systems. Two time series predictive Artificial Neural Network (ANN) algorithms: Multi-Layer Perceptron (MLP) and Long-Short-Term-Memory (LSTM) were also employed to predict the sensor response and to compare.