Variational autoencoder analysis gas sensor array on the preservation process of contaminated mussel shells (Mytilus edulis)

Mussel shells is a macro zoobenthos that lives on soft substrates in the mud (infauna) and is classified as a bivalve. This research detects formalin in mussel shells utilizing an Electronic Nose comprised of gas sensor's array. The samples used were formalin mussel shells with several concentrations from 100 ppm to 500 ppm with the addition of 100 ppm. The research was conducted using six sensors with a sampling time of 120 s. The output voltage from each sensor is then clustered based on principal component analysis and classified using several techniques, which are support vector machine, decision tree and random forest. We demonstrate that all classifiers have an accuracy of 1. The phenomenon occurs because all feature representations can produce enough information to classify data. Principal component analysis achieves the best score in preserving the local structure. PCA can keep an average of 33% nearest data in the same neighbourhood. While variational autoencoder can keep 14% nearest data in the same neighbour, and autoencoder can keep 8% nearest data in the same area.