Predicting drug dissolution profiles with an ensemble of boosted neural networks: a time series approach.

Applicability of an ensemble of Elman networks with boosting to drug dissolution profile predictions is investigated. Modifications of AdaBoost that enables its use in regression tasks are explained. Two real data sets comprising in vitro dissolution profiles of matrix-controlled-release theophylline pellets are employed to assess the effectiveness of the proposed system. Statistical evaluation and comparison of the results are performed. This work positively demonstrates the potentials of the proposed system for predicting desired drug dissolution characteristics in pharmaceutical product formulation tasks.