Shuffled Frog-Leaping and Weighted Cosine Similarity for Drift Correction in Gas Sensors

Artificial Olfactory Systems (AOS) mimic the Biological Olfaction (BO) using sensors and artificial intelligence algorithms. The performance of an AOS is based on the sensitivity and the selectivity of the sensors against the sensed odors. Long term sensors drift is a major problem that brings distortion in the sensitivity of the sensors. Traditional methods to compensate drift, like sensors replacement, successive recalibration or domain transformations are either expensive or not always feasible to use. In order to overcome the issues related to traditional approaches, a novel method based on a sequential classification approach is proposed in this paper, with following contributions: (i) A Tree Structured Cosine Similarity (TSCS) based classification approach is proposed to handle long term sensors drift. (ii) The classifier is embedded within a memetic metaheuristic Shuffled Frog Leaping Optimization (SFLO) approach to optimize the features space. (iii) The proposed approach works as a combinatorial optimization problem that recursively reduces the number of features and increases the classification accuracy of the system, while compensating the drift. (iv) Only median values of optimized features are enough to train the classifier which reduces the computational cost and the memory requirement of the classifier. (v) The proposed approach is purely based on the feature subset selection process for drift compensation without requiring any data samples from target domain or system recalibration, making it suitable for the real life applications. The proposed approach is compared with the existing state-of-the-art approaches using an extensive experimental dataset and a significant increase in the classification accuracy is observed.