Enhancing the Operation of 13 GOOD Market

Project Details


Fall 2015


I Chun Chao, Yi Chun Chuang, Sherry Wu, Chia Li Chien





a. Market Introduction
Hsinchu 13 good market has been established in April, 2014, and opens every Saturday. Most of the customers are the households living nearby, so the market will hold events for family as well.
The market has their own promotion channel “Facebook fan page”. In January, 2016, the
accumulative page “likes” is up to 4,950 in over 1 years.
Since this is almost the most important and the only way to publish their news and
information to the customers. We wondering if there is any possible way to make good use
of the fan page data to make a better promotion of the market.

b. Business Problem and Background
We took the manager of Hsinchu 13 GOOD market as our client.
The aim is to help the manager adapt a proper promotion strategy for the venders who
really need. We try to classify the revenue fraction of each vender by whether it is above
3% or not. By doing that, the market manager can know if she should take extra
promotional action to certain vender instead of spending extra cost on those who already
got much attention. We hope we can save not only the advertising cost, but also the
working hour of the staff who is in charge of FB fan page.
On every week, we can take a look on the revenue fraction of the vender we did promotion
for to see its growth. If the fraction is growing, we can say that we are success.

c. Data Description
We get the sales data from 13 good market, including revenue, the product categories, and
the activities they held. Besides market information, we also get the fan page data from
Facebook insights. Then, we collect the temperature and other environmental data as
factors as well. Finally, we combine all the dataset into one sheet.
We take “vender on each market day” as one record and the main output variable is if it
need promotion (Yes / No).

d. Analytics Solution
According to the specialties of our data, we create a lot dummies first. Being terrified by so
many columns, we run PCA. Then, we try to use several analytic methods, such like
Logistic Regression, Naïve Bayes and x validation in Rapid Miner. In order to find some
additional explanation for our data, we also take a shot at visualization.

e. Conclusions and Recommendation
Due to what we see from all the result, we recommend the market manager to take other
more promotional channels and plan for the fall promotion activities. Although we did get
a very large dataset at the end of the semester, we still get a good model to help the
manager decide which vender should be promoted. And we hope that we can help our
client save cost on advertisement and labor through a well performing model.

Application Area: