| Abstract [eng] |
The purpose of this work is to create non liner regression model with optimal number of coefficients and optimal values of these coefficients, and predict liquor concentration having values of amperimetric data. The main task of the work: • To use Semi-supervised learning algorithm while creating the model (Kai Yu, Volker Tresp [4]) • To apply P.L. Bartlet [1] idea, that in non-linear regression model the value of the weights is more important than number of these weights. • To create mathematical model that could calculate optimal non-liner regression’ weights using stochastic search, and could determine the concentration of liquor according biosensors response curve While creating the model, we met model optimization problem. The optimal model means the optimal number of model’ weights and the values of weights with whom the model best predicts the concentration of the liquor. We used step wise regression method to determine the optimal number of weights. For the calculations we choose non-liner regressions sigmoid function according P.L. Bartlet [1] work results. We also implemented semi supervised data analysis method that was chosen based on Volker Tresp [3] work results. These methods are widely applied in the mathematical modeling and information technologies. What is more, we also applied partial component analysis and reduced the data sets of response curve, without loosing significant information about these data. The parameters , of the model we chosen in experimental way while observing the models MSE. The results showed that this model predicts the values of the concentration very good with appropriate parameters and appropriate number of regressors. |
