Gas Sensor Drift Compensation by Ensemble of Classifiers Using Extreme Learning Machine

Abstract

The step by step deteriorated chemically, in the sensor response over long time, when the external conditions are constant, is known as sensor drift. Sensors drift is one of the most grave worsening of bothering gas sensors as in electronic nose. In this respect, drift mitigation is an important issue. Traditional techniques are expensive and difficult because they need to regularly recalibrate referred gases or continually provide data labeling. In this work, software based approach for drift compensation is proposed by not changing the hardware components, which in turn reduces the cost. To verify our proposed approach, gas sensors drift dataset is retrieved from the UCI machine learning repository. The dataset has been collected for three years of six different volatile organic compounds, under tightly controlled operating conditions using an array of 16 metal-oxide gas sensors. In this work, extreme learning machine has been implemented, for very fast and efficient performance. Extreme learing machine is used both as single standalone classifier and ensemble base classifier. It is shown out of these two methods ensemble technique has handled the problem of drift more efficiently and reasonably improved the classification accuracy to compensate the challenge and helps in pattern recognition in compensating the sensor drift.