The variable being predicted is discrete in that presumably only whole/integer numbers of TVs can be sold (but if the numbers being sold are large enough then it may be possible to approximate it to a continuous distribution, e.g. to a Normal distribution, with a continuity correction).
I wouldn't call this a classification problem, because the outcome variable is not categorical. So I agree with lnathan that it is a regression problem, not a classification problem.
As a starting point for this type of data, if there were lots of low counts (0's and 1's etc.), I would use a general linear model with a Poisson distribution, or with a negative binomial distribution if needed. Or if the counts were large (generally, lambda>10) then I would approximate to a Normal distribution.
I would tend to use logistic regression for cases where the outcomes are categorical rather than numeric. In its simplest form this is where the outcome is binary (e.g. 'Yes'/'No').