In this study, our aim was to estimate the adsorption potential of three families of aerogels: nanocellulose (NC), chitosan (CS), and graphene (G) oxide-based aerogels. The emphasized efficiency to seek here concerns oil and organic contaminant removal. In order to achieve this goal, principal component analysis (PCA) was used as a data mining tool. PCA showed hidden patterns that were not possible to seek by the bi-dimensional conventional perspective. In fact, higher total variance was scored in this study compared with previous findings (an increase of nearly 15%). Different approaches and data pre-treatments have provided different findings for PCA. When the whole dataset was taken into consideration, PCA was able to reveal the discrepancy between nanocellulose-based aerogel from one part and chitosan-based and graphene-based aerogels from another part. In order to overcome the bias yielded by the outliers and to probably increase the degree of representativeness, a separation of individuals was adopted. This approach allowed an increase in the total variance of the PCA approach from 64.02% (for the whole dataset) to 69.42% (outliers excluded dataset) and 79.82% (outliers only dataset). This reveals the effectiveness of the followed approach and the high bias yielded from the outliers.
Keywords: aerogel; machine learning; oil removal; principal component analysis; sustainability; unsupervised learning.