Predicting Customers' Purchase Tendency for Precision Marketing to Increase ​Migo​’s Revenue

Project Details


Fall 2019


Shao-En Lin, Hsuan-Yu Chen, Tzu-Hung Lin, Wan-jung Tsai





Migo is a multinational business that seeks to bridge the digital gap in emerging markets for consumers. They provide an online-to-offline platform that allows consumers to go to the nearest Migo WiFi zone to download videos and watch offline without buffering anytime. In this project, we worked on the data mining task to predict customer’s purchase tendency for the next month to increase Migo’s revenue.

Business Goal

The business goal of Migo is to increase the number of paid users. They would like to identify who is the high-value customer, and adjust their marketing campaign monthly to achieve steady growth. For example, they might give special discount for loyal or high-value customer. Thus, a data mining solution that could assist Migo to find their potential high-value consumers is expected. We believe our data mining prediction will benefit marketing and business develop team in Migo for making managerial decision in the case of precision marketing.

Data Mining Goal

To achieve business goal, our objective is to predict whether a user will purchase or not in the upcoming month by the data of the past 2 weeks. As we consulted with Migo’s specialist, we determined the time interval as one month since Migo is used to validate their campaign results monthly.


We have the dataset of transaction, engagement, download and title from 2017-09-01 to 2017-12-31. We transformed the raw data into 23 input variables, and the outcome will be binary variable showing whether a customer will purchase or not.

Method and Evaluation

We applied naive bayes, logistic regression, decision tree and random forest. The benchmark we considered to use is precision, since it’s more important for investors to know who will really purchase within our prediction. We chose ROC curve as to evaluate our prediction, the result turns out that random forest has the best performance.


This is an ongoing analysis as new transactions are received by old users and as new users start buying Migo products. The data need to be updated and run on a monthly basis.


To sum up, we suggest Migo to consider the followings:
1. For those who will purchase in the next month, offer them an invitation code to invite others visit Migo and get discount in return.
2. PisoMigo, 3-day UNLI Pass, Red MC starter pack and movie are the most popular
products and content type, further campaign could be targeting on these products. 

3. The most important features recommended by several model is how many times watched on the title instead of how much money customer spend, so the engagement behavior could be a much more important factor compare to spending.

Application Area: