Impact of Oversampling on the Classification of Readily Biodegradable Materials

With an eye on controlling environment pollution, many countries have formulated rules to keep a check on the use of non-biodegradable chemicals by industries. Computational methods have an important role to play in predicting biodegradability as experimental methods are pretty expensive. This paper investigated the effectiveness of dimensionality reduction methods (principal component analysis, kernel principal component analysis, sparse random projection and gaussian random projection) and classifiers such as random forest and support vector machine in classification of biodegradable compounds. As the data set was imbalanced, experiments were also carried out using data oversampled by Synthetic Minority Oversampling Technique (SMOTE). The best classification performance was obtained by support vector machine classifier which used data balanced by SMOTE and dimensionally reduced using sparse random projection. This model was more effective in predicting chemical biodegradability when compared to other related works.