Guide to Churn Prediction: Part 3 — Descriptive statistical analysis
Jahnavi C.
Growth
TLDR
Data exploration and analysis is a timeconsuming process. We’ll continue to explore the Telco Customer Churn dataset to get a brief overview of data by using the descriptive statistical analysis method.
Outline
Recap
Before we begin
Exploratory Data Analysis
Descriptive statistical analysis
Conclusion
Recap
In part 2 of the series,
Guide to Churn Prediction, we explored the
Telco Customer Churn
dataset to identify dependent and independent features, find missing values, and understand their mechanisms.
Before we begin
Let’s understand the
2
common terms that we often hear when dealing with data.
Population data
Source: Giphy
Data collected from the
entire population
is known as
population
data. The population is dynamic (it changes continuously), so it’s practically impossible to gather population data.
Sample data
Source: Giphy
Data collected from a
subset
or a specific
group
of population is known as
sample
data.
Note
: The Telco Customer Churn dataset, which we are using in this series, is
sample
data.
Step 4: Exploratory Data Analysis (EDA) contd..
To explore and analyze data, we’ll perform EDA in
3
different methods shown below:
Descriptive statistical analysis:
The process of using statistical techniques (like mean) to summarize the sample data. This analysis gives us a brief
overview
of the sample data.
Descriptive graphical analysis
: The process of analyzing the data with the aid of graphs. This analysis provides us with
indepth knowledge
of the sample data.
Inferential statistical analysis:
The process of making inferences about the population data using statistical methods (like hypothesis testing, etc.). In a nutshell, this analysis helps us
make assumptions
about
population data.
Assumptions are made based on insights gained while performing descriptive and graphical analysis on features of the
sample data.
Now, let’s go through the
descriptive statistical analysis
method in detail.
Source: Giphy
Descriptive statistical analysis, also known as
descriptive statistics or summary statistics,
describes and summarizes the sample data. It gives us information about the
measures of central tendency
(mean, median, mode/frequency) and
measures of dispersion
(standard deviation, range) of the sample data.
Let’s use descriptive statistical analysis to answer the following questions:
1. What are the summary statistics of the numerical features?
2. What are the summary statistics of categorical features?
Let’s start with importing the necessary libraries and loading the cleaned dataset. Check out the link to
part 1to see how we cleaned the dataset.
1
2
3
4
import pandas as pd
df = pd.read_csv('cleaned_dataset.csv')
df
Cleaned dataset
Create 2 datasets,
df_num
and
df_cat
, with
df_num
containing only
numerical
features and
df_cat
containing only
categorical
features. It’s much easier to do descriptive and graphical analysis when the features are separated.
Let's start by checking the data types of features using the
dtypes
function.
1
df.dtypes
Types of features
Based on the data types, separate the features and create 2 new datasets as shown in the code below.
Numerical Features
Categorical features
Let’s use the
describe()
function to display the summary statistics of the features. This function displays the count, mean, median, standard deviation, 25%, 50%, 75%, min, and max values of each feature in the dataset.
count
displays the number of
nonnull records/rows. This value will be the same as the number of records/rows in the dataset if there are no null values.
mean
displays the average value of the data.
std
(standard deviation) value tells us approximately how far each data point/record is from the mean.
min
and
max
displays the minimum and maximum values, respectively.
25%, 50% (median), 75%, and 100%
displays the values at that percentile.
For example,
25%
displays the
value at the 25th percentile.
68% of data lies in (
mean — std and mean+std)
range.
95% of data lies in (
mean — 2*std and mean+2*std)
range.
99.7% of data lies in (
mean — 3*std and mean+3*std)
range.
If the
mean
is equal to the
median
, then the distribution is
normal
.
If the median is
greater
than the mean, then the
distribution
of the data is
right

skewed
.
If the median is
less
than the mean, then the
distribution
of the data is
left

skewed
.
1
df_num.describe()
Summary statistics of numerical features
For example, let’s understand the summary statistics of the “Total Charges” column.
The count of nonnull records is 7032, which is not the same as the number of records in the dataset. So, this indicates that there are null values in the “Total Charges” column.
On average, a customer paid approximately $2,283 at the end of the quarter.
The maximum amount paid by a customer at the end of a quarter is approximately $8,684.
At the end of the quarter, 50% of the customers paid approximately $1,397 or less than that.
The mean value (2,283.3) is greater than the median value (1,397.47), so the distribution of the data is rightskewed.
Now, let’s see what interesting insights we can gain from the table.
More than
25%
of the customers
stopped
using their services by the end of the quarter.
On average, a customer has been with the company for
32
months and paid around $
2,283
in total.
Customers who have been using their services for
more than 55 months
are more likely to cancel their services by the end of the quarter.
Let’s use the
describe(exclude=[‘int64’,’float64’])
function to display the summary statistics of the features. This function displays the count, unique, top, and most frequent values of categorical features.
count
displays the number of
nonnull records/rows. This value will be the same as the number of records/rows in the dataset if there are no null values.
unique
displays the total number of unique values, i.e., number of categories in a feature.
top
displays
the first record/row in the feature.
freq
displays
the mode.
1
df_cat.describe(exclude=['int64, 'float64'])
Summary statistics of categorical features
For example, let’s understand the summary statistics of the “Churn Label” column.
There are no null values.
There are 2 categories in the “Churn Label” column.
The first record has the label “No”.
One category appears 5174 times, indicating that 5174 records belong to 1 category and the remaining records (7043–5174 = 1869) to the other.
All customers are from the state of California.
There are 2 types of phone services and 3 types of internet services.
All customers have the option to have up to 3 multiple lines.
There are 3 types of contracts and we may be able to identify a customer who is more likely to churn based on the contract type.
The categories in the “Churn Label” column are not evenly distributed, so we say that the data in this column is
imbalanced
.
Try analyzing the columns and see if you can get more interesting insights from the data.
Conclusion
Consider a scenario where your client has given you a large dataset and has asked you to immediately provide some information based on the given data.
Source: GIPHY
In such circumstances, descriptive statistical analysis can be performed quickly, and a brief overview of the data can be provided to the client.
That’s it for this blog. Next in this series, we’ll see how to explore and analyze the data using the
descriptive
graphical analysis
method
and see what interesting insights we can gain through the graphs.
Thanks for reading!!
Start building for free
No need for a credit card to get started.
Trying out Mage to build ranking models won’t cost a cent.
No need for a credit card to get started. Trying out Mage to build ranking models won’t cost a cent.