The parameters of the Hindmarsh-Rose model are evaluated by measuring the output of the model and estimating the derivatives of the output since direct derivatives measurement leads to poor conditionality of the problem. A "Dirty-Derivative" filter of the third order is used to evaluate them, which gives a smoothed estimate of derivatives. The parameter estimation algorithm is applied to real EEG data and demonstrates convergence of tuning parameters to some values. The obtained parameter estimates are used as features for epileptic seizure classification problem.