What makes couples stay together is a very interesting topic. The group not only describe the modeling process in the report but also interpreted the result very well.
The group first tried to use Relationship Quality as Y, but later found out it is seriously skewed upward. Most couple would rate their relationship as successful thus making Relationship Quality not a objective feature to predict. Instead, the group looked at whether a couple still stay together after six years. This is a more concrete definition of success than Relationship Quality. It shows that for prediction models, having a concrete definition of Y is very important. Ratio, nominal and interval variables should be considered first before ordinal variables.
This project also raised the issue of scaling X. Sometimes, we will have some features with exceptionally large values. In this case, to make the models more predictable, we should normalize large value features.
In the conclusion section, the issue that different group of people have different standard for successful relationship was raised. For example, some people from traditional families will choose to stay in marriage even if the relationship has stranded. To tackle this issue, we can either find a more universal definition of successful relationship or divide people into different groups and fit each group with a more detailed model.
What makes couples stay together is a very interesting topic. The group not only describe the modeling process in the report but also interpreted the result very well.
The group first tried to use Relationship Quality as Y, but later found out it is seriously skewed upward. Most couple would rate their relationship as successful thus making Relationship Quality not a objective feature to predict. Instead, the group looked at whether a couple still stay together after six years. This is a more concrete definition of success than Relationship Quality. It shows that for prediction models, having a concrete definition of Y is very important. Ratio, nominal and interval variables should be considered first before ordinal variables.
This project also raised the issue of scaling X. Sometimes, we will have some features with exceptionally large values. In this case, to make the models more predictable, we should normalize large value features.
In the conclusion section, the issue that different group of people have different standard for successful relationship was raised. For example, some people from traditional families will choose to stay in marriage even if the relationship has stranded. To tackle this issue, we can either find a more universal definition of successful relationship or divide people into different groups and fit each group with a more detailed model.