Gas classification using binary decision tree classifier

Abstract

Gas classification with an array of sensors is challenging for real life applications due to the limited amount of available training data of gases. Different pattern recognition algorithms are successfully used for gases identification, but their performance is degraded when the training and testing of these algorithms is done with different concentrations data. In this paper, we are using a binary decision tree approach for gas classification, and we are considering difference in the sensitivities of the sensors in every pair of a multi-sensor array as an input attribute for the tree. Suitable pairs of sensors are found by exploring their capability to split the available gases data samples at the decision node of the tree into two branches. A distance metric is used to select a single sensor pair in the case of more than one pair of sensors for the gases distribution at the decision node. The selected pairs of sensors learned during the training phase at the decision nodes are applied on the test data vectors. The effectiveness of our algorithm is successfully verified on the acquired data set with an array of seven metal oxide gas sensors for five different gases.