Preprocessing of electronic nose data by independent component analysis

Arrays of chemical sensors are affected by different disturbances. They are due, in part to the environment and in part to the non-ideality of the system. Since sensors cannot be completely insulated from the surrounding environment the fluctuation of ambient parameters (e.g. temperature and relative humidity) may be so large to strongly limit the resolution of the sensor array. In this work, Independent Component Analysis (ICA) are introduced as a preprocessing method aimed at reducing the influence of spurious disturbances on the meaningful part of the data. The method is shown on a practical experiment aimed at distinguishing the ripeness stage of apples measured at different room temperatures and relative humidity. The benefits in term of improved classification will be illustrated and discussed.