Normalization of data Generally the data is normalized with 0 mean and unit variance. Is it mandatory to use this method? Can I normalize data by simply dividing it with the mean of data (which is another normalization technique in data mining)?
Thanks in advance,
 A: What normalizing does is usually eliminating the effect of units. For example if you have a measurement in cm for example 1cm then it is .01m and .001km. Now assume you have another measurement for example weight, simple 1km, then it is .001ton and so on. Now the goal of analysing is comparing these two measurements. Obviously you cannot compare weight and height directly, then we have to find a way to normalize data in order to remove the measurements and make them comparable. On the other hand if you use very small units then it is difficult to cope with computations. Moreover, normalizing needs to provides you with some simplicities, for example saying $\frac{x-\mu}{\sigma}$ is preferable in statistics because in many cases the target distribution tends to Gaussian that this king of normalizations helps a lot. Another example is linear regression, if you normalize data to have a mean zero, then you basically removed the intercept and simplified the computation process a lot.
To sum up, you can use any kind of normalization you like but need to keep in mind that normalizing must give you something more than what you have already.
PS: there is not a consensus among statistician about normalizing. Some of them recommend that and some of them are against that.
