PURPOSE: The purpose of this study was to evaluate the efficiency of a neuro-fuzzy logic-based methodology to model poorly soluble drug formulations and predict the development of the particle size that has been proven to be an important factor for long-term stability. METHODS: An adaptive neuro-fuzzy inference system was used to model the natural structures within the data and construct a set of fuzzy rules that can subsequently used as a predictive tool. The model was implemented in Matlab 6.5 and trained using 75% of an experimental data set. Subsequently, the model was evaluated and tested using the remaining 25%, and the predicted values of the particle size were compared to the ones from the experimental data. The produced adaptive neuro-fuzzy inference system-based model consisted of four inputs, i.e., acetone, propylene glycol, POE-5 phytosterol (BPS-5), and hydroxypropylmethylcellulose 90SH-50, with four membership functions each. Moreover, 256 fuzzy rules were employed in the model structure. RESULTS: Model training resulted in a root mean square error of 4.5 x 10(-3), whereas model testing proved its highly predictive efficiency, achieving a correlation coefficient of 0.99 between the actual and the predicted values of the particle size (mean diameter). CONCLUSIONS: Neuro-fuzzy modeling has been proven to be a realistic and promising tool for predicting the particle size of drug formulations with an easy and fast way, after proper training and testing.
PURPOSE: The purpose of this study was to evaluate the efficiency of a neuro-fuzzy logic-based methodology to model poorly soluble drug formulations and predict the development of the particle size that has been proven to be an important factor for long-term stability. METHODS: An adaptive neuro-fuzzy inference system was used to model the natural structures within the data and construct a set of fuzzy rules that can subsequently used as a predictive tool. The model was implemented in Matlab 6.5 and trained using 75% of an experimental data set. Subsequently, the model was evaluated and tested using the remaining 25%, and the predicted values of the particle size were compared to the ones from the experimental data. The produced adaptive neuro-fuzzy inference system-based model consisted of four inputs, i.e., acetone, propylene glycol, POE-5 phytosterol (BPS-5), and hydroxypropylmethylcellulose 90SH-50, with four membership functions each. Moreover, 256 fuzzy rules were employed in the model structure. RESULTS: Model training resulted in a root mean square error of 4.5 x 10(-3), whereas model testing proved its highly predictive efficiency, achieving a correlation coefficient of 0.99 between the actual and the predicted values of the particle size (mean diameter). CONCLUSIONS: Neuro-fuzzy modeling has been proven to be a realistic and promising tool for predicting the particle size of drug formulations with an easy and fast way, after proper training and testing.