Target Pregnancy Predictor

Case Study 4 Related Facts Companies such as Target are looking for new ways for customers to shop at that location. Through market research, they learned that a person 's life is about to change purchasing habits. This is the most important of these times when a woman has her first child because the baby needs the product. By carefully examining the purchasing habits of 25 items, Target can determine the possibility that a woman will become pregnant and a baby will expire.

For example, in response to consumer purchasing behavior, Target not only can accurately predict pregnancy but also accurately predict delivery dates. This is not rocket science - the goal is to know that you purchased a pregnancy test (doh!), Then an odorless body lotion, and then purchased zinc and magnesium supplements. They compiled a list of 25 products listing the target pregnancy prediction scores. When you were found, boom: diaper coupons arrived at your doorstep within time. This prediction is primarily based on customer survey, but the self-learning algorithm can go further and predict before predicting pregnancy. ) These algorithms can more historically utilize structured or unstructured data (such as the number of steps). Age, sharing of social media), find patterns from thousands of examples, learn and predict results (such as pregnancy) with a certain probability. This is called data science, especially self study.

Not all products are targeted to everyone. Apps that track pregnancy are not targeted to teenage boys. However, even if you do not have users or viewers who are not targeting specific content, feedback from users other than the users you are targeting can help you identify the marginal situation you may not consider. Maybe this teen boy needs to use a pregnancy tracker to help the sisters without his girlfriend or mobile phone - you never have to be someone outside the target audience to be your active user of your product I do not know.

A collection is a collection of predictors summarized to obtain a final prediction (for example, the average of all predictions). The reason for using collections is that many different predictors are trying to predict that the same target variable will show better performance than a single predictor alone. Integration technology is further divided into bagging and boosting. Normally, each model uses a random sub-sample / data boot loader, so there is little difference between all models. Select each observation and replace it with the input of each model. Therefore, each model is observed differently depending on the boot process. In this method, many unrelated learners need to create a final model, so reducing the variance reduces the error. The whole example of bagging is a random forest model.