Simple data exploration in warranty service

Introduction

One of my friends asked me methods to present her specific data. This post came as a result. The data was randomly generated. No trending was found, it was just some basic thought to visualize data in a colorful and interactive way.

Warranty service report

1. Data
2. Descriptive statistic
3. Go deeper?
4. Suggestions

Data

• Randomly created
• From 01/01/2015 to 31/12/2015
• 70000 records
• 4 brand names
• 4 product types
• 3 staffs

Notice: The data was generated in a rush, therefore it looks funny and seems meaningless as every factor looked as the same.

Sample data

register_day	release_day	keep_time	user	brand	product_type
2015-07-11	2015-07-18	7	user 1	Sony	smart phone
2015-04-20	2015-04-30	10	user 1	Sony	desktop
2015-08-07	2015-08-24	17	user 1	Microsoft	desktop
2015-06-23	2015-07-07	14	user 2	Microsoft	desktop
2015-07-26	2015-08-04	9	user 3	Microsoft	smart phone
2015-05-23	2015-06-05	13	user 1	Microsoft	smart phone
2015-09-28	2015-10-10	12	user 3	Microsoft	smart phone
2015-02-21	2015-03-02	9	user 1	Asus	notebook
2015-08-09	2015-08-22	13	user 1	Microsoft	laptop
2015-03-13	2015-03-29	16	user 2	Dell	laptop

Register and release activities

How many product were register?

How many product was handle by user?

How long a user handle a product in avarage?

Why we need to look into distribution of an indicator?

The indicator we focus on here is keep_time (time a product was kept in warranty service). People tend to use avarage of keep_time in most cases but this indicator is always sensitive to outlier. Thus, we use a box plot to have a broaden view:

• Median value.

• Common values of keep_time, limit by first quantile and the third quantile. 50% percent of cases has the keep_time falls into it. Our expected standard (for example we want to keep_time below 14 days) was best when it was higher this range.

• Outliers: these were the values much higher than the average value. Usually we need to go deeper to figure out what was going on there. There were also many low outliers when the keep_time is too small, we should ignore it because it might be the bugs too simple

• Go here for a easy explanation of box plot

Distribution of how long a brand name was kept?

Distribution of how long a user fix a product

Distribution of how long a product type was kept

Distribution of how long a product type was kept consider to brand name

Distribution of how long a product type was kept consider to user

Go deeper?

• Was there a time that we register too many products could lead to overheat for warranty team?
• Abnormal changes in warranty team (interupt, new people)?
• ANOVA test.

Suggestion

• Not sure if we could go further with just some random data.