Project: Machine Learning for Raman Spectroscopy

Figure: Sowoidnich, Kay. (2013). Optics for food inspection – Non-destructive detection methods using Raman spectroscopy. Qioptiq Optolines. 32. 14-16.
In this project, we aim to apply machine learning techniques to Raman Spectroscopy data and provide a understanding of chemical properties from a data analytical perspective. Our input data are a set of simulated Raman spectra for a set of molecules and corresponding molecular features, e.g. Double Bond Equivalents (DBE), number of Bridgehead Carbon Atoms. The baseline problem is to use extracted features from Raman spectra to make predictions on molecular features for corresponding molecules. A more
challenging problem is to demonstrate how different machine learning classification algorithms (e.g. random forest, artificial neural network (ANN), and ensembles of ANN) react to Raman spectra with different levels of added noise. An even more challenging problem is to determine if machine learning classification algorithms are able to predict molecules that are not available during training.

Researchers:   Dr. Timko, Chem. Engineering and   Dr. Paffenroth, Math Sciences & Data Science, PhD Student  Wenjing Li