h2>Dating : 10 Traits of Successful Daters
A logistic regression classifier to predict someone’s popularity in dating
Dating may have changed drastically over the last decade, but the desire to connect with each other certainly hasn’t, nor has it ever been easy to play the field.
It’s an interesting question to ponder, “what makes someone popular on the dating market?” Is it gender, looks, intelligence, or something else? And I’d like to dig deeper into that question with a speed dating dataset from kaggle.
In this speed dating experiment, the researchers surveyed 500 participants from 21 speed dating sessions involving, each with similar numbers of male and female participants. The survey included questions pre-, peri-, and post-event, resulting in 100 variables, from basic information such as age, gender, ethnicity, hobbies, to one’s assessment of their own personality and attractiveness, as well as their impression of other participants during the event, and follow-ups after the event. While some information was missing, this was a relatively comprehensive dataset.
More than match or no match, I’m hoping to predict how someone fares in dating without prior information on what they meet and if they end up with any matches, I’m only looking at data collected before each event.
I picked the following variables as a start. These include identity-related information, hobbies, self-assessments, and their expectations from the event.
Whether someone is ‘successful’ in a speed dating session is determined by the ‘dec_o’ column, which stands for decisions from the opposite sex. It indicates the number of people from the opposite sex that are interested in seeing them again. In modern dating lingo, this is the equivalent of swiping right on someone.
I’m defining ‘success’ as being liked by more than50% of the participants, of course, the threshold can be changed to run different models.
Based on that information, I created a new variable ‘popularity’ with values ‘popular’ and ‘less popular’.
Out of curiosity, I wanted to see both popularity distribution and gender distribution among those who are popular. It’s interesting to note that while being a female does seem to make a difference, it is definitely not the sole determinator of popularity, as conventional wisdom might suggest.
A correlation heatmap also revealed that there are some moderate correlations and I decided to drop a few variables, including ‘concert’, ‘museums’, and ‘music’.
Out of the three models, logistic regression, random forest, and XGBoost, logistic regression seems to give me the best overall results, based on validation accuracy and F1-score. And with further feature selection, I was able to build a model with a 70% accuracy. This means I can predict whether or not someone will have more than half of the people ‘swipe right’ on them with 70% accuracy, given the provided information.
Currently, I’m working on improving the model, and also making it adaptable to various thresholds so we can analyze those who are highly successful in dating or perhaps find out why someone struggles to find a date.
Lastly, if by now you start feeling a little intimated by all the competition in the dating market, here’s a gentle reminder: at the end of the day, being popular has nothing to do with finding true love!
