Group14 proposal

Motivation and Objectives

Nowadays, all industries in the world are facing fierce competition. With the development of telecom technology and social media, the Telco companies play more and more important role in the society. There are growing number of wireless carriers in the world. The U.S has four main wireless carriers and lots of little wireless carriers. It is no surprise that the companies in this industry face very fierce competition. Since this condition, the most significant problem for these organizations are customer remaining. As we know, companies from these industries often have customer service department. Their target is that winning back clients who is churn. Because it is generally acknowledged that recovering long-term customers can be worth much more to a company than acquiring new customers.

In order to understand more directly the main factors that affect customer churn and better maintain the relationship with customers, relevant models will be built so as to select and visualize important variables. Lastly, we will present the comparison among models towards Recall, Accuracy, Precision and F1 score and evaluate the performances of different models.

Critique of Existing Visualization

This radar chart has all the variables from the dataset to be presented in a single graph. We can acknowledge that by different series (“0” as “No”, “1” as “Yes”), each variable their number of customers. However, even though this visualization has revealed plenty of valuable information, such as unbalanced data in several variables with a fair enough graph, some questions can be spotted from the graph. For instance, too many variables have presented on the chart at the same time. The best way to solve this problem is to use feature engineering, then to draw a new graph to visualize better the influence of different factors to churn or not churn.

Data Description

We collect the dataset from IBM Community. This dataset contains five spreadsheets. They contain the information about the demographics, location, population, services and status about customers. Demographic is the information about customers’ gender, age range, and if they have partners and dependents. Location is the information about customers’ detail location such as country, city. Status is the information about customers’ status of churn and the reason about churn. There are 7043 entity instances in the dataset. Each customer is identified by Customer_ID column. There are 42 columns with 40 attributes. Customers who left within the last month is the column named Churn_Value. The churn customers are recorded as 1 and the non-churn customers are recorded as 0.

Data Fields	Description	Example	Datatype
Customer ID	Customer ID	7590-VHVEG	Numeric
gender	Whether the customer is a male or a female	Female	Binary
SeniorCitizen	Whether the customer is a senior citizen or not (1, 0)	0	Binary
Partner	Whether the customer has a partner or not (Yes, No)	Yes	Binary
tenure	Number of months the customer has stayed with the company	1	Numeric
PhoneService	Whether the customer has a phone service or not (Yes, No)	No	Binary
MultipleLines	Whether the customer has multiple lines or not (Yes, No, No phone service)	No phone service	Categorical
InternetService	Customer’s internet service provider (DSL, Fiber optic, No)	DSL	Categorical
OnlineSecurity	Whether the customer has online security or not (Yes, No, No internet service)	No	Categorical
OnlineBackup	Whether the customer has online backup or not (Yes, No, No internet service)	No	Categorical
DeviceProtection	Whether the customer has device protection or not (Yes, No, No internet service)	No	Categorical
TechSupport	Whether the customer has tech support or not (Yes, No, No internet service)	No	Categorical
StreamingTV	Whether the customer has streaming TV or not (Yes, No, No internet service)	No	Categorical
StreamingMovies	Whether the customer has streaming movies or not (Yes, No, No internet service)	No	Categorical
Contract	The contract term of the customer (Month-to-month, One year, Two year)	Month-to-month	Categorical
PaperlessBilling	Whether the customer has paperless billing or not (Yes, No)	Yes	Binary
aymentMethod	The customer’s payment method (Electronic check, Mailed check, Bank transfer (automatic), Credit card (automatic))	Electronic check	Categorical
MonthlyCharges	The amount charged to the customer monthly	29.85	Numeric
TotalCharges	The total amount charged to the customer	29.85	Numeric
Churn	Whether the customer churned or not (Yes or No)	No	Binary

Methodology and Approach

In Feature engineering, we will generate variables from the previous ones and compose multiple features together, after which we will separate churn and not churn customer and separate categorical and numerical columns. The main point of this is selecting effective variables which would result in customer attrition. Apart from that, we may transform multi value variables to the dummy variable in the last step of this stage, we can obtain the variable summary like below:

Secondly, a correlation matrix of this model will be visualized to present the relationship among different variables and primarily understand the influence of each variables.

We intend to build four different models and visualize them and their performances: (1) In Logistic regression model performance, confusion matrix and receiver operating characteristic will be visualized to evaluate the performance of the Logistic regression model. Besides, we can get the variable importance while comparing all variables by bar chart. Furthermore, obtaining the appropriate threshold for logistic regression will also be visualized to understand what is beneath the model. (2) Synthetic minority oversampling Technique will be applied to build the advanced logistic regression, and we will try to form user portrait using the most important features which are selected from the model. (3) The decision tree will be generated refer to the results of the feature score and compare the GINI coefficient to measure the degree of inequality of the distribution. And we intend to use high-score categorical features to make a ternary plot to visualize the distribution of two groups. (4) In order to control over-fitting and improve the predictive accuracy, we will also build and visualize Random Forest Classifier and compare different trees. Lastly, we will visualize the comparison among four models towards Recall, Accuracy, Precision and F1 score and compare the performances of different models.

Team Members

• Qiu Yang
• Shi Yawen
• Zhu Keyu