Sensor Array Optimization for Mobile Electronic Nose: Wavelet Transform and Filter Based Feature Selection Approach

Abstract

Mobile Electronic Nose (MoLen) is a prospective concept for Sensing as a Service (S2aaS) development. Furthermore, gas sensor array is a substantial part of MoLen. This work treats about two issues related with gas sensor array. First, commonly used resistive sensor e.g. MOS (Metal-Oxide Semiconductor) gas sensor is highly contaminated with noise. Second, a poor combination of sensor array leads to features redundancy issue. These problems cause significant performance degradation on classifier. It will get worse if the classifier runs on S2aaS environment. To deal with these issues, this study proposes the robust sensor array optimization method based on Wavelet Transform to handle the noise and the modified Fast Correlation-Based Filter (FCBF) to find the best combination of sensor array with minimizing feature redundancy. This study has the following contribution: i) reducing the noise from gas sensor array that generated irrelevant data; ii) finding the best sensor array for beef quality classification to improve the quality of input to classifier. The experimental results show that the proposed method successfully reduces the noise power at maximum 14.41% and it is able to determine the best combination of sensors in sensor array with the 16% of improvement of General Resolution Factor (GRF)that is associated with larger classification rate.