I did a frequency analysis table, for around 23 previous studies then I choose the most five frequency factors, the examiner asked to Justify why I chose only five variables from that table?
Dear Israa Najeeb Zaidan ,
Randomly choose a subset of features is usually unfair.
if the studies did not identify such features as very important, you better use more of them.
If there are many studies using such features, then you may write they are the most common.
The appropriate number of variables depends on the type of process you are studying. Without describing the subject matter it is not possible to comment on the number of variables.
Indeed, without specifying the problem, this is not easy to answer.
But if anyway a general answer: You can test different numbers of variables and investigate how that influence your result. If, say, 5 variables and 10 variables gives the same result, then using Ockham's razor, you can conclude that 5 is preferred, as it uses less variables (degrees of freedom).
Regarding sound is is pretty easy to argument for using few parameters. We know that loudness matters, e.g. LAeq or LAeq max [dB], reverb is a class of sound, T30 [s], if the sound is related to speech intelligibility, then the Speech Transmission Index, STI [1] may be useful (but it is inversely dependent on the T30 and the background noise). In many cases single number parameters is used in requirements and measurement standards.
However if more detail in the spectra is to be revealed more numbers my be useful, like octave band spectra or one third octave bands. If you aim to do a statistical analysis of the recorded spectra, an anova or a x-y plot of all possible combinations may be feasible and useful, but the limits for saying that a correlation is significant becomes more and more strict the more hypothesis tests you do.
If you are more specific about the problem, we may be able to help you better.
- Badges
- Science topic