Multiple classifier combination technique for sensor drift compensation using ANN & KNN

Abstract

Drift in sensors, mainly in chemical or gas sensors is an unavoidable problem that introduces shift in feature values in the dataset. This makes sample classification and identification process more challenging over time in olfactory machines. The generation of uncertain chemical sensor drift is long term degradation of the sensor properties and no matter what they are made of, how expensive they are, or how accurate. To deal with this problem a multiple classifier approach using artificial neural network (ANN) and k nearest neighbour (KNN) is proposed here and tested with the gas sensor array drift dataset which is retrieved from UCI machine learning repository. At first the extensive dataset is processed using principal component analysis (PCA) for visualization of underlying clusters. Then in order to supervise the problem and counteract its effect, drift compensation techniques using multiple classifiers using ANN (BP-MLP)& KNN have been formulated. Finally, a comparative study on the efficiency of ensemble of classifier for the single standalone classifier in terms of average classification accuracy is evaluated. The results clearly indicate the superiority of multiple classifier approach which not only improves the classifier performance but also compensate with sensor drift concept without replacing the physical sensor for long term use.