Classification of VOC Vapors Using Machine Learning Algorithms

Detection of volatile organic compound (VOC) vapors, which are known to have carcinogenic effects, is extremely important and necessary in many areas. In this work, the sensing properties of a cobalt phthalocyanine (CoPc) thin film at six different VOC vapors (methanol, ethanol, butanol, isopropyl alcohol, acetone, and ammonia) concentrations from 50 to 450 ppm are investigated and it is observed that the interaction between the VOC vapors and CoPc surface is not selective. It is shown that using machine learning algorithms the present sensor, which is poorly selective, can be transformed into a more efficient one with better detection ability. As a feature, 10 seconds of raw responses taken from steady state region are used without any additional processing technique. Among classification algorithms, k-nearest neighbor (KNN) reaches the highest accuracy of 97.1%. The selected feature is also compared with classical steady state response feature. Classification results indicate that the feature based on 10 seconds of raw responses taken from steady state region is much better than that based on classical steady state response feature.