1

I'm starting to learn about data science. I have sample data including different features run over several people, for each person n times. these features represent the fluency of the participants in some languages. they are including the number of correct words per sentence, the average number of the correct words in total, the average of speaking duration, the average of pauses in each turn, etc. So for features such as the number of correct words, the higher the score the better, and there is a feature such as average pauses which the lower the score the better.

I want todecide at the end which features are more significant and worth keeping and which features can be ignored.

I read about the weighted composite scores, and as far as I understood the main step is figuring out the coefficients (weights). but I didn't understand how these coefficients should be calculated. some cases are based on the analysis intuition, as I saw in some examples, but it can be so subjective.

Subhash C. Davar
  • 578
  • 4
  • 18
Lili
  • 11
  • 3

2 Answers2

1

This is a tentative answer.

One can try to determine first the relative importance of each feature (eg by Factor Analysis or Principal Component Analysis, ..)

Once the more significant features have been identified (or guessed perhaps), then one can try combinations of scores with these features.

  1. Take simply the average as a composite score.
  2. Do a linear regression and use the coefficients as the weights of the composite score.
Nikos M.
  • 2,301
  • 1
  • 6
  • 11
  • after applying PCA and selecting PC1 and PC2, how can I identify significant features based on PC1 and PC2? – Lili Apr 05 '21 at 21:53
  • ignificance in PCA is based on the variance of each featuire, that is the singular value. Usualy in PCA the feature vectors are sorted in descending order of variance or signular value, so the first are the ones which carry the greatest information, thus in a sense more significant – Nikos M. Apr 06 '21 at 15:35
0

Factors/features that determine the fluency significantly can be ascertained with the help of multiple regression analysis. Certain proposed features may have minor or insignificant impact on fluency of participants . These can be dropped from your proposed regression model for a general predictive-modeling. Some authors seem to call regression coefficients as weights or beta weights. Regression coefficients are a distinct measure and reflect effect of particular variable/factor on a specific dependent measure.

Subhash C. Davar
  • 578
  • 4
  • 18