Jan-17-2022, 05:14 PM
(This post was last modified: Jan-17-2022, 05:15 PM by data_wiz_5.)
Hi All,
I’m currently trying to find out an algorithmic approach to find the most in demand profile group for a set of survey data, this is example of the dataset I’m working with:
Survey |Demand |Age |Gender
----------------------------------------
1 |5 |[18,19,20] |[1,2]
2 |10 |[19,20,21] |[1,2]
3 |5 |[20,21,22] |[1,2]
• There are 3 surveys to be done
• Survey 1 needs 5 people to complete it, and the requirements are that they are aged 18,19 or 20 and male/female
So, from visually looking at the data the most in demand profile is people aged 20, followed by those aged 19 or 21.
However, I need to be able to run this on a larger dataset, so I’m looking for an algorithmic approach to find the same conclusion as above.
What I’ve tried so far is:
• Creating unique combinations of all possible answers and counting demand by combination – this unfortunately doesn’t scale very well as surveys can have multiple questions with 30+ options, so the number of combinations becomes unusable.
• I’ve thought about using a decision tree, which given that it uses a gini index, could be very useful to determine which questions are the most significant drivers of demand. However, I’ve been unable to find a way to fit the model given that my dataset stores a range of possible options.
Other assumptions to consider:
• The example above is a simplification with only 2 questions, there are many more questions in the dataset
• Some will have a larger range of answers (e.g., 30+)
• Not all surveys would have the same questions (i.e., some will ask more/less questions than others)
Any help, thoughts or general guidance would be greatly appreciated!
I’m currently trying to find out an algorithmic approach to find the most in demand profile group for a set of survey data, this is example of the dataset I’m working with:
Survey |Demand |Age |Gender
----------------------------------------
1 |5 |[18,19,20] |[1,2]
2 |10 |[19,20,21] |[1,2]
3 |5 |[20,21,22] |[1,2]
• There are 3 surveys to be done
• Survey 1 needs 5 people to complete it, and the requirements are that they are aged 18,19 or 20 and male/female
So, from visually looking at the data the most in demand profile is people aged 20, followed by those aged 19 or 21.
However, I need to be able to run this on a larger dataset, so I’m looking for an algorithmic approach to find the same conclusion as above.
What I’ve tried so far is:
• Creating unique combinations of all possible answers and counting demand by combination – this unfortunately doesn’t scale very well as surveys can have multiple questions with 30+ options, so the number of combinations becomes unusable.
• I’ve thought about using a decision tree, which given that it uses a gini index, could be very useful to determine which questions are the most significant drivers of demand. However, I’ve been unable to find a way to fit the model given that my dataset stores a range of possible options.
Other assumptions to consider:
• The example above is a simplification with only 2 questions, there are many more questions in the dataset
• Some will have a larger range of answers (e.g., 30+)
• Not all surveys would have the same questions (i.e., some will ask more/less questions than others)
Any help, thoughts or general guidance would be greatly appreciated!