bachelorThesis
Desenvolvimento de modelos de calibração e classificação de resíduos de medicamentos em leite por meio de espectroscopia de infravermelho NIR e quimiometria
Fecha
2019-12-12Registro en:
MARIA, Ellen Diane. Desenvolvimento de modelos de calibração e classificação de resíduos de medicamentos em leite por meio de espectroscopia de infravermelho NIR e quimiometria. 2019. Trabalho de Conclusão de Curso (Bacharelado em Química) - Universidade Tecnológica Federal do Paraná, Pato Branco, 2019.
Autor
Maria, Ellen Diane
Resumen
The presence of drug residues in milk, such as Avermectins, causes several problems for human health and damages to the dairy industries, being necessary to perform analyzes to detect the presence of these constituents. Most analyzes performed for this purpose require a large sample volume, long run time and generate chemical waste. Thus, this work presented the proposal to develop multivariate calibration models for detection and quantification of Abamectin and Doramectin in raw milk by the Near Infrared (NIR) technique associated with Partial Least Squares Regression (PLS). In addition, this work aimed to construct classification models to differentiate Abamectin and Doramectin by NIR spectroscopy combined with Support Vector Machine (SVM). Raw milk samples were doped separately with different concentrations of Abamectin and Doramectin, and after tampering the NIR spectra were generated from the entire sample set on the FT-NIR MPA spectrophotometer. PLS models were constructed and different pretreatments were tested on the spectral data set. Four models using PLS were built using ChemoStat software, two for each type of drug and the quality of the models was evaluated using the values of R2, RMSEC, RMSEP and RMSECV. Weka software generated the classification model for the two drugs, Abamectin and Doramectin. The quality of the classification model was assessed by the merit figures: accuracy, precision, positive hit rate and area under the curve (ROC curve). The SVM classification model presented excellent results both in the calibration and validation of the model, as all merit figures presented adequate values for this modeling. The built PLS models obtained R2 values higher than 0.90 and low RMSEC values in the calibration, however the validation results were not as satisfactory for any of the drugs, due to their R2 lower than 0.55, which shows the need for adjustments and improvements in the proposed technique. The development of this work shows that the use of the NIR spectroscopic technique has excellent advantages and that it can be applied in the industrial sector, since its execution takes place in a short analysis time, has a moderate cost and minimal waste generation. Moreover, from the improvement of the proposed methods, their application can be performed together, since both methods, one for quantification and another for classification of medicines in milk use the same spectral data, making possible the classification of these types of drugs. residual caresses in milk.