Why to remove skewness from the data? I am a beginner in statistics and I read an article which said "Linear algorithms love normally distributed data". I wanted to know why do we have to transform the variable having skewness. 
 A: If you transform skewed data to make it symmetric, and then fit it to a symmetric distribution (e.g., the normal distribution) that is implicitly the same as just fitting the raw data to a skewed distribution in the first place.  In fact, any skewed distribution can be written as a symmetric distribution on a non-linear transformation of the underlying variable, so this method is just one way to characterise a skewed distribution.
Note that this is not really a matter of removing skewness from the data.  Rather, we are making another transformed set of data values where the skewness is removed and then modelling this with a symmetric distribution.  Of course, if we want to go back and make conclusions about the values on the original scale we will need to transform back again.  In this respect, the transform is really there just to simplify our statement of the model we are fitting to the data.
An example: The log-normal distribution is a positively skewed distribution.  It can be formed as a normal distribution on the logarithms of an underlying set of values.  So in some problems you will get a set of positively skewed non-negative data, and you might be able to model these by taking logarithms of the data values and then fitting these transformed values to a normal distribution.  If you do this, that is equivalent to just fitting the original (untransformed) values directly to a log-normal distribution.
