2. Introduction
Background
Reviewed consumer complaints data from the Consumer Financial Protection
Bureau (CFPB), an independent agency of the U.S. government, responsible
for consumer financial protection.
Analyzed complaints for Bank of America between 2012 to 2015 to better
understand if the data had any actionable insights that an organization could
use.
About Bank of America
Bank of America is an American multinational banking and financial services
corporation headquartered in Charlotte.
It is the second largest bank holding company in U. S. by assets.
Products- consumer banking, corporate banking, finance and insurance,
investment banking, mortgage loans, private banking, private equity, wealth
management, credit cards.
3. Analysis Techniques
Business Objective 1: Predict and forecast Responses to
Consumer Complaints to better manage resources (monetary
and people)
Build a Decision Tree to predict the Company's response to
consumer complaints
Build a Forecasting model to forecast Complaints/timing of
Expenses
Business Objective 2: Manage Brand image from Online Reviews
Text Analytics on the complaint descriptions from Twitter
5. About the Dataset
The data contains information regarding complaints
received for different financial organizations.
The dataset contains the fields - Complaint ID,
Product, Sub-product, Issue, Sub-Issue, State, ZIP
Code, Submitted via, Date received, Date sent to
company, Company response, Timely response,
Consumer disputed.
We focus on complaints with Date Received between
1/1/2012 and 12/31/2015. There is a total of
516696 complaints in this time period.
6. Complaints by Product
Highest number of complaints are for Mortgage product (176,975), followed by Debt Collection
(92,446), Credit Reporting (82,598) and Credit card (62,326).
7. Top 10 Companies by Number of
Complaints
Bank of America receives the highest number of complaints (53,876) followed by Wells Fargo
and JPMorgan Chase.
8. Company Responses to Credit Card
Complaints
Closed with Explanation
(35,412) and Closed with
Monetary Relief (12,823) are
the top two company responses
for Credit Card Complaints.
9. Exploratory data analysis: Bank of America
Bank of America is the fourth most complained company .
Total number of complaints for bank of America is 6846
Top 10 Most Complained about
Credit Card Companies
10. Bank of America: Complaints by Product
Majority of the complaints are for Mortgage (35,284) followed by Bank Account or service
(9,809)
11. Bank of America: Issues by Product
The top issues for Bank of America Bank account or service are Account opening, closing, or
management (4,120) and Deposits and Withdrawals (2,684).
12. Bank of America: Issues by Product
The top issues for Bank of America credit cards are Billing disputes (1,106) and Other(671)
followed by closing/cancelling account (633).
13. Bank of America: Complaints across
States
Highest number of complaints for Bank of America were received from
California(10,271), followed by Florida (6,128)followed by New
York.
14. Bank of America: Complaints across
States for Credit Card Product
Highest number of complaints for Bank of America Credit Card
product were received from California (966), Florida (685) and New
York(610)
16. Analysis Techniques
Business Objective 1: Predict and forecast Responses to
Consumer Complaints to better manage resources (monetary
and people)
Build a Decision Tree to predict the Company's response to
consumer complaints
Build a Forecasting model to forecast Complaints/timing of
Expenses
Business Objective 2: Manage Brand image from Online Reviews
Text Analytics on the complaint descriptions from twitter
17. Predictive Modeling: Decision Tree
Decision Tree to predict the Company's response to consumer
complaints
Target variable - Company response
Input variables - Product, Issue, Submitted via, Timely response,
Consumer disputed,
Built using rpart package
We divide data into training(900 observation), test datasets (304
observation). Training data is used for building model and test
data is used to test the selected model.
20. Predictive Modeling: Decision Tree
Validation
Test the decision tree model accuracy.
While classifying in closed with explanation the
accuracy is 90%.
21. Time-series Forecasting model
Build a Forecasting model to forecast Complaints/timing of
Expenses
Analyzed seasonality in complaints (specific quarter of times of the
year, peaks in complaints ) and interesting trends.
Built using R forecast package and ARIMA forecasting models
Input variables - Number of Complaints against Date Received
Time interval – quarter
22. Time-series Forecasting model
The default model selected by the system is ARIMA Model with
lower MAPE 9.39. This means that, on average, there is a 9.39%
difference between the forecast and actual values. The forecast
model is usually the model with the best fit (lowest MAPE) among
all of the models considered for the series
24. Time-series Forecasting model
Long term trend shows that complaints are reducing over time.
It also shows some seasonality in complaints in APR Month.
25. Text Analytics on Online Reviews
Text Analytics on the complaint descriptions from Twitter
Used text clustering technique to understand natural groupings
and capture any interesting insights from the complaints
description text
Extracted 64 reviews for Bank Of America from Twitter using
R(TwitterR package) software
Built using R.
Sentiment analysis of twitter data to align customer reviews.
Cluster algorithm - Number of clusters – 3
27. Text Analytics on Online Reviews
Sentiment analysis on the data and scores assigned and classify
them in to positive, negative ,neutral.
28. Text Analytics on Online Reviews
Cluster 1 looks like Customer Service experience as the reviews
talk about help ,bad,good,dispute,cancel,agree etc.
Cluster 2 looks like Billing, fee and payment issues as the reviews
talk about start ,late ,close ,pay ,look ,payment etc.
Cluster 3 looks like Account and fund processing as the reviews
contain terms such as fund, check, deposit, process, etc.