Combining wavelength importance ranking to the random forest classifier to analyze multiclass spectral data.

Near Infrared (NIR) is a type of vibrational spectroscopy widely used in different areas to characterize substances. NIR datasets are comprised of absorbance measures on a range of wavelengths (λ). Typically noisy and correlated, the use of such datasets tend to compromise the performance of several statistical techniques; one way to overcome that is to select portions of the spectra in which wavelengths are more informative. In this paper we investigate the performance of the Random Forest (RF) classifier associated with several wavelength importance ranking approaches on the task of classifying product samples into categories, such as quality levels or authenticity. Our propositions are tested using six NIR datasets comprised of two or more classes of food and pharmaceutical products, as well as illegal drugs. Our proposed classification model, an integration of the χ2 ranking score and the RF classifier, substantially reduced the number of wavelengths in the dataset, while increasing the classification accuracy when compared to the use of complete datasets. Our propositions also presented good performance when compared to competing methods available in the literature.

Identifying the most relevant tablet regions in the image detection of counterfeit medicines.

This paper proposes a novel image-based approach to detect counterfeit medicines and identify the most relevant regions of the tablet in the task of classification. Images of medicine tablets undergo an initial pre-processing step which (i) removes the background to find the region of interest, (ii) clusters individual pixels into super-pixels, and (iii) extracts features containing color and texture information. The classification relying on Support Vector Machine (SVM) defines the class the respective image will be inserted into. The task of identifying the relevant regions of the tablets for counterfeiting detection is performed using the concept of support vectors, generating a heat map that indicates the regions that contribute the most to the classification purpose. Two datasets containing images of authentic and counterfeit tablets of Cialis and Viagra were used to validate our propositions, achieving correct classification rates of 100% on both datasets. Regarding the task of identifying the most relevant regions, our proposition outperformed the traditional LIME (Local Interpretable Model-agnostic Explanations) method by yielding more robust explanations.