The primary experimentation with electronic nose on black tea has established the fact that the instrument has the potential to be used in the shop floor on a regular basis for instrumental evaluation of the quality of black tea. But its introduction to the user industry, in its present form, is difficult due to non-availability of some generic computational model capable of predicting tea taster like scores for the quality of tea samples. This being a new technology, the tea industry personnel are reluctant as usual and expect that the electronic nose will generate the results with accuracy at par, if not better, than the experienced tea tasters. Evolution of such generic and universal computational model is not possible without elaborate experimentation with the new instrument over several seasons and at different tea producing places. And such data collection exercise would involve a significant number of tea tasters, huge number of tea samples with variations of (a) location of plantation, (b) clonal characteristics and (c) seasonal varieties.
Such complex data collection plan is important since quality of tea varies with the place and season in which it is produced. For example, in north and north-east India, the tea leaves are plucked
in five sessions over a year, called as flushes and they are termed as the first flush, the second flush, the rain flush, the autumn flush and the winter flush. Out of these five flushes, the quality
of tea produced in the second and the autumn flushes is usually better than the tea produced in the other three flushes in the same garden. If the pattern classifier is trained with the dataset available
in one garden during a particular flush, it is unlikely that the same model may give correct results when subjected to data from a different garden or in a different flush. It is thus expected that, for
evolution of a truly universal co-relation model between electronic nose signatures and tea tasters’ scores, an appropriate incremental learning algorithm would be the most preferred pattern classifier.