Customer Analytics Case Study Report
Workshop 3
Explore & Analyse with Spotfire
1
Predicting Customer Churn
Using Spotfire to predict whether a customer (mobile services) will leave the company.
We will:
Explore the data to identify pertinent features.
Perform logistic regression on a subset of variables.
Develop a classification tree on a subset of variables.
Analyse the outcomes and comment of the results.
Examine implications for further studies.
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
Predicting Customer Churn
IMPORT
Load the file ‘cell2cell_churn’ into Spotfire.
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
Predicting Customer Churn
IMPORT
The following is a summary of field descriptions for select fields.
| Field | DataType | Description |
| Churn | Categorical | Did the customer switch to a different provider? |
| MonthlyRevenue | Numeric | Average monthly spend on phone services. |
| MonthlyMinutes | Numeric | Average number of minutes of call per month. |
| DroppedCalls | Numeric | Average number of calls that lost connection per month. |
| CustomerCareCalls | Numeric | Average number of calls to customer service per month. |
| ActiveSubs | Numeric | Number of services the customer subscribes to. |
| Handsets | Numeric | Number of handsets the customer has had over time. |
| HandsetModels | Numeric | Number of different handset models the customer has had over time. |
| CurrentEquipmentDays | Numeric | Number of days the customer has had their current handset. |
| HandsetWebCapable | Binary | Is the handset able to browse the web. |
| OwnsComputer | Binary | Whether the customer owns a computer. |
| HasCrediCard | Binary | Whether the customer has a credit card. |
| Credit Rating | Categorical | A scale of 1 to 7, 1 being highest quality and 7 being lowest quality. |
| Income Group | Categorical | A scale of 1 to 9, 1 being the lowest income stratum and 9 being the highest. |
| Prism Code | Categorical | A reclassification of the user’s residential post code into one of Rural, Suburban, Town or Other. |
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
Predicting Customer Churn
EXPLORE
Create a stacked barchart to compare where customers live (PrizmCode) and their socioeconomic status (IncomeGroup)…
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
Predicting Customer Churn
EXPLORE
Create a stacked barchart to compare creditworthiness (CreditRating) and their socioeconomic status (IncomeGroup)…
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
Predicting Customer Churn
EXPLORE
Create a stacked barchart to compare credit card possession (HasCreditCard) and their socioeconomic status (IncomeGroup)…
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
Predicting Customer Churn
EXPLORE
Heatmap – This produces a matrix that can show the average monthly revenue for customers as defined by their credit rating and income level.
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
Predicting Customer Churn
EXPLORE
Heatmap – We can also examine the topology for monthly revenue by credit card possession and credit rating.
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
Predicting Customer Churn
EXPLORE
Waterfall Chart – What is the monthly revenue contribution by income group?
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
Predicting Customer Churn
EXPLORE
Waterfall Chart – What is the monthly revenue contribution by credit rating segments?
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
Predicting Customer Churn
EXPLORE
KPI Chart – average monthly revenue and handsets per customer by income group.
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
Predicting Customer Churn
ANALYSE – LOGISTIC REGRESSION
Select ‘Classification modelling’ from the Tools menu.
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
Give your model a name
Add a short comment
Choose ‘Logistic Regression’
Select ‘Churn’ as the response column
Choose what you believe are good predictors of churn.
Use the ‘Add’ button and select variables of interest to add to the model.
Select OK when ready
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
Predicting Customer Churn
ANALYSE – LOGISTIC REGRESSION
Akaike Information Criterion: A measure of overall model efficacy.
Predictor coefficient: The size of the response per unit change in the predicator.
P-value (significance): How significant is the independent (predictor) variable.
Akaike Information Criterion
Predictors of
customer churn
Probability of
significance
Predictor
coefficients
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
https://www.sciencedirect.com/topics/medicine-and-dentistry/akaike-information-criterion
Predicting Customer Churn
Summary – LOGISTIC REGRESSION
Akaike Information Criterion:
A measure of overall model efficacy.
Used to compare against other models.
The lower the better.
Predictor coefficient:
The size of the response per unit change in the predicator.
E.g.: ‘CreditRating5-Good’ has coefficient -0.16, suggesting customers with good crediting ratings tend to leave this company.
E.g.: ‘CreditRating2-Low’ has coefficient +0.25, suggesting customers with substandard crediting ratings tend to stay with this company.
P-value (significance):
How significant is the independent (predictor) variable.
At 95% confidence, we look for p ≤ 0.05
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
https://www.sciencedirect.com/topics/medicine-and-dentistry/akaike-information-criterion
Predicting Customer Churn
ANALYSE – CLASSIFIATION TREE
Select ‘Classification modelling’ from the Tools menu.
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
Give your model a name
Add a short comment
Choose ‘Classification Tree’
Select ‘Churn’ as the response column
Choose what you believe are good predictors of churn.
Use the ‘Add’ button and select variables of interest to add to the model.
Select OK when ready
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
Predicting Customer Churn
ANALYSE – CLASSIFIATION TREE
Notable outputs are:
ROC curve – Indicative of model efficacy, ideally it should bend.
Predicted outcome probabilities – a probability for each response, sums to 1.
Note: classification trees do not produce AIC statistics so ROC becomes basis for comparison.
Churn probabilities:
p(NO) = 71.2%
p(YES) = 28.8%
ROC curve of a good model
ROC curve of our model indicates a poor fit.
Workshop 3
-Slides 17-22 "Using analytics to predict customer churn": there is brief mention here of machine learning; feel free to rework it and add technical examples of decision trees and logistic regression if you have anything fitting (6 slides)
https://www.kaggle.com/kmalit/bank-customer-churn-prediction
https://www.sciencedirect.com/topics/medicine-and-dentistry/akaike-information-criterion
Workshop 4
Recency, Frequency, Monetary Model with Python
Workshop 4
-Slides 11-20 are on RFM (Recency, Frequency, Monetisation) Analysis - just background info for you
-Slide 24: FYI, Sanjeev's suggestion about doing an RFM using Python (which not all students can do)
20
Predicting Customer Churn
Two examples in Python:
Both require either:
(1) the demo-ing lecturer to install a python environment on their computer or
(2) use the analytics server to run the code
https://medium.com/capillary-data-science/rfm-analysis-an-effective-customer-segmentation-technique-using-python-58804480d232
https://towardsdatascience.com/recency-frequency-monetary-model-with-python-and-how-sephora-uses-it-to-optimize-their-google-d6a0707c5f17
Predicting Customer Churn
xxx
https://medium.com/capillary-data-science/rfm-analysis-an-effective-customer-segmentation-technique-using-python-58804480d232
https://towardsdatascience.com/recency-frequency-monetary-model-with-python-and-how-sephora-uses-it-to-optimize-their-google-d6a0707c5f17
Predicting Customer Churn
xxx
https://medium.com/capillary-data-science/rfm-analysis-an-effective-customer-segmentation-technique-using-python-58804480d232
https://towardsdatascience.com/recency-frequency-monetary-model-with-python-and-how-sephora-uses-it-to-optimize-their-google-d6a0707c5f17
Predicting Customer Churn
xxx
https://medium.com/capillary-data-science/rfm-analysis-an-effective-customer-segmentation-technique-using-python-58804480d232
https://towardsdatascience.com/recency-frequency-monetary-model-with-python-and-how-sephora-uses-it-to-optimize-their-google-d6a0707c5f17