Predicting the Type of Nanostructure Using Data Mining Techniques and Multinomial Logistic Regression

Nanotechnology and nanomaterials have a promised future in different aspects of modern life that involve medicine, environment, space, energy, electronics, security, and many others. While the applications of nanomaterials seem to be limitless, new challenges are also being posed. With regard to the type of one-dimensional nanostructure of Cadmium Selenide (CdSe), there are three possible morphologies presented: nanosaws, nanowires, and nanobelts. Since the synthesis of these morphologies are by trial and error, our goal in this paper is to use statistical and data mining techniques to predict the type of CdSe nanostructure. The methods used for prediction are: a multinomial logistic regression, a support vector machine, and a random forest. The results are compared using two statistical indices: sensitivity and specificity, and the factors that influence the possible nanostructure are identified. Based on the results, data mining techniques showed to be a better fit for prediction comparing to the multinomial logistic regression model. We also identify the levels of these factors that maximize the proportions of nanosaws, nanowires, and nanobelts. : Nanotechnology; CdSe nanostructure; Prediction; Data Mining; Multinomial Logistic Regression.