Creating a composite score from dataset with no target variable

Question

I have a dataset that includes 6 variables about prospective sales opportunities (probability of closing, days until expected close, age of opportunity, etc.). 2 of the columns are categorical and 4 are continuous. I am looking to create a composite score of each opportunity based on these variables.

I have tried manually assigning scores to each column based on arbitrary ranges, but the composite scores ended up being too similar. I want to optimize the scoring for each variable. Is there a way to do this through machine learning? I want to find out which variables are the most important, or "stand out" the most from the rest of the pack. I also would like to see if any variables are redundant.

Answer 1

If you dont have the outcome of the observation, it is not possible to find which variables are more important (important to predict what ? the « what » is the target value).

I encourage you to gather some records (20 times the number of features is a good starting point) with targets so the algorithm can understand what features are relevant to the specific outcome you want.

However it is already possible to calculate correlations between variables and do some data analysis to understand how much variance each feature contains.