Dear Mr Manan, sensors full selective is very difficult. There is almost always some interference. Except sensors based on antibodies (at moment). In the sixties years, potentiometric sensors were named specific ("full selective"). After, it was discovered that were selectives. During five decades, the research in this area was to discover sensors more selectives for the specie of interest. Amperometric glucose sensors is an example. Electronic tongue uses also potentiometric sensors, but in this case is not desirable "full" or the highest selectivity. In the last decade, with advance in the computational techiques, multivariate calibration was viable. However, it is just possible to separate the measure signals by the multivariate algorithms if there is some multicolinearity. Then, for electronic tongue, selectivity is not desirable. I think that the answer for your question is just due the historical evolution of the concept and an adaptation to manner as measure shall be made. However, this is my opinion based on a historic vision. I can be wrong.

Dear Mr Manan, a high degree of multicollinearity also difficult the multivariate algorithm to perform the matrix inversion required for computing the regression coefficients. Moreover, if the sensor is very selective, there isn't response signal for all species. This question is very interesting because this seems be a contradition.

Hi Dr. Fernandes. Thank you for your answers. Through my experience in working with fusion of electronic nose and electronic tongue, I discover few things regarding the nature of both sensors. I thought because the nature of e-nose that is fully selective and e-tongue that is partially selective, both sensors inherit high multicollinerity and fair multicollinerity, respectively. From my previous works in doing classification for fusion of e-nose and e-tongue data, the results show that the fusion scheme is able to increase the classification performance. But when talking about the classification performance of single modality, we know that e-nose always perform better than e-tongue. I believed that these scenarios have to do with the nature of fully and partially selective sensor. Do you agree with that statement? Thanks.

Dear Mr Masnan, I don't know. I have a bit experience with potentiometric sensors and multivariate calibration, but the fusion between electronic nose and tongue not. I though that electronic nose had low selectivity, once it used sensors based on metal oxides that changed its resistence when a gas is adsorbed on its surface. For me, any volatile molecule produced a change in the electrical resistance. But this was in the first sensors. The comparison is dificult because the sensors used in electronic nose are different than ones used in electronic tongue. Could be the mathematical algorithm instead the sensors? Did you try to use others algorithms based on neural network or genetic algorithm for fusion of both sensors? Electronic tongue is an area very interesting for me because uses potentiometric sensors. I will make download your last article. I need to know more to have an opinion grounded.

Dear Dr Fernandes, for your information, I am using Cyranose320 - an e-nose based on conducting polymer, while for the e-tongue - potentiometric sensors. Actually, I was trying to figure out the relationship between selectivity of sensors with multicollinearity of sensors. I need to know the relationship before I finish my phd (statistics). The focus of my study is classification of multi sensor data fusion based on feature selection. This is my personal opinion regarding the issue. Please correct me if I am wrong. e-tongue is an ion specific sensor which means that each working electrode will detect specific chemical. If two electrodes detect the same chemical, it may be because that particular electrodes detect different chemicals but that particular chemicals share the same ions. And for the reference electrode, it will not have any relationship with other working electrode. This scenario can be seen from the scatter matrix plot of data from e-tongue. And it is very obvious that the correlations among working electrodes are not really high (usually at most is 0.7), except for the reference electrode (no corelation with any other working electrode). As for the e-nose, it is known to be fully selective but detects non specific volatile organic compounds (VOCs). Because the array sensors from e-nose are able to detects the same VOCs, it increases the interference of among sensor array, and this is the main factor for high correlation among array sensors of e-nose. And the high correlation may be between 2 or more sensors due to the nature of nonspecific detection of VOCs. Usually the correlations among array sensors are in a range of 0.8 to 0.9 ++. Any comment? Thanks!

Dear Mr Manan, your work is very interesting. I understood that you explained. But, there is a contradition between the chemical and statistical concept. Potentiometric sensors are selectives no more specifics. The reference electrode is just used to maintain the electrical potential from Earth, constant. Without it, indicator electrode or ion selective electrode (ISE) would not produce a response. Therefore, the reference electrode is just important for measuring electronic. It is possible an ISE to respond for another chemical specie, but this is considered an undesirable interference. The selectivity coefficients measure how much a sensor responds for one specie. Hence, potentiometric sensors must be classified as "full selectives" and not "partially selectives". On the order hand, e-nose uses resistive or piezoeletric ( I think) sensors. Once sensors in e-nose respond to any volatile specie, they aren't selectives, thus they must be classified as "partially selectives". You are correct when say that potentiometric sensors are ion "specific" (more correct is selective!) sensors, but you are wrong to describe one only sensor used in e-nose as "fully selective". But I am talking in just a sensor. E-nose and e-tongue are a sensor matrix. Maybe the combinatorial analysis of the array of sensors produces an inversion. Therefore, from chemical view point for just a sensor, potentiometric sensors are "full selective sensors" and resistive or piezoeletric sensors are "non (partially?!) selective sensors". However, from statistical view point using array of sensors, e-tongue are "partially selective sensors" and e-nose are "full selective sensors". I can be wrong, but I read that in e-tongue is not convenient to use sensor extremely selectives. For example, potassium ion selective electrode uses valynomicin as chemical compound in its membrane. This electrode is very selective for potassium ions, and respond almost 10000 times more for potassium than another ion (sodium, calcium, for example) . This phenomenon is not convenient to use in e-tongue. Thus, in e-tongue, the potentiometric sensors are purposely chosen to have low selectivity.

You can see this article: E-Nose and e-Tongue combination for improved recognition of fruit juice samples, Food Chemistry, Volume 150, 1 May 2014, Pages 246–253

It may be give you information about the interest of E-Nose and e-Tongue combination.

Dear Aguir, thank you very much for your suggestion. I will take a look the paper.

Have you guys considered using Total Least Squares, Partial Least Squares, or ridge regression? Normally these are used under multicollinearity.

I have read all of the responses but none seem to answer the real question. What does selectivity mean? Can someone answer the question? I am also interested in finding out.

The selectivity of a sensor gas corresponding to certain features (eg. Material of the active layer) to determine if the device is capable of responding to a particular group of volatile compounds (analytes). In electronic noses this concept is shown in the discrimination of a data set, where the sensors work on the principle of overlapping sensitivities.