UCLA

Speed dating is a popular and fast way to meet new people and find life partner in nowadays society. Four professors from Columbia university did the speed dating experiment from 2002-2004 and I use their data-set in this paper to answer the research question: what are the gender differences on selecting opposite sex partner from speed dating event, and can we eventually predict people’s decisions. In this speed dating experiment, every participant had a chance to meet a new person from opposite sex just through a 4-minute conversation. They collected everyone’s basic information such as gender, race, age and so on. Before the speed dating event they also collected participants’ hobbies, expectation about opposite sex, and what kind of person themselves are. During the event, each participants would also value how they think their partner is. In this paper, I did basic data analysis to explore gender difference and other useful information about different people’s preference on opposite sex. I try to predict males’ final decision, if they like the female who they just met, using either all information I have before or after the speed dating event. The base model I use is logistic regression model, and I improved the model by step-wise variable selection. The compared models are decision tree model, random forest model and XGBoost model. I separated the whole data set into 80% training data and 20% testing data to avoid over-fitting. The best model I finally have is XGBoost model. It has a 82.4% precision on testing data-set based on all the information we have after the speed dating event, and still a 70.2% precision on testing data-set even we only use all information before two people never actually met on the speed dating event. So we can believe that we have the ability to discover the secret of love with modern machine learning algorithms if we have enough information.