190 likes | 541 Vues
Customer Satisfaction/Loyalty. Turna Koksal. Goal. Characterize the customer of a bank Customer satisfaction Customer loyalty Relationship between satisfaction and loyalty. Domain. Collection of answers given to survey questions by customers 6500 customer records 174 attributes .
E N D
Customer Satisfaction/Loyalty Turna Koksal
Goal • Characterize the customer of a bank • Customer satisfaction • Customer loyalty • Relationship between satisfaction and loyalty
Domain • Collection of answers given to survey questions by customers • 6500 customer records • 174 attributes
Method • Association rules • Relationships among items in dataset • WEKA • Apriori algorithm
Implementation • Data cleaning • MS Excel • Clean data (Derived attributes) • Attribute selection • WEKA • Information gain algorithm • Top 15 attributes (>0.15)
Implementation (cont.) • Data transformation • Transform attributes into nominal values • Attribute values 1 to 7 and 99 • Group into 4: • {1,2,3} 1 • {4,5} 2 • {6,7} 3 • {99} 4 • Divide into 6 groups • Attribute QTA : A,B,C,D,E,F
Implementation (cont.) • Data transformation • Divide data into training (70%) & testing (30%) • Transform training file into .arff format • Rule properties • Generate 20 rules for each group • Minimum confidence: 0.8 • Minimum support: 0.45
Dataset Rule • Q2_01=3 ==> Q2_03=3 • Information on site arranged logically = {strongly agree, agree} • Trust bank to protect privacy & confidential info = {strongly agree, agree} Accuracy: 60.86%
Group A Rule • Q20=3 ==> Q2_03=3 • How satisfied with online services {extremely satisfied, very satisfied} • Trust bank to protect privacy & confidential info {strongly agree, agree} Accuracy: 53.83%
Group B Rule • Q2_01=3 ==> Q2_03=3 • Information on site arranged logically {strongly agree, agree} • Trust bank to protect privacy & confidential info {strongly agree, agree} Accuracy:62.98 %
Group C Rule • Q48=3 ==> Q49_02=3 • Overall satisfaction with bank {extremely satisfied, very satisfied} • Remain customer {extremely likely, very likely} Accuracy: 49.23%
Group D Rule • Q2_01=3 ==> Q2_03=3 • Information on site arranged logically {strongly agree, agree} • Trust bank to protect privacy & confidential info {strongly agree, agree} Accuracy: 58.12%
Group E Rule • Q2_01=3 Q2_06=3 ==> Q2_03=3 • Information on site arranged logically {strongly agree, agree} • bank.com helps me take charge of my finances {strongly agree, agree} • Trust bank to protect privacy & confidential info {strongly agree, agree} Accuracy: 54.29% &
Group F Rule • Q1_01=3 ==> Q49_02=3 • Web site overall {extremely satisfied, very satisfied} • Remain customer {extremely likely, very likely} Accuracy: 51.22%
Next-Steps • Try different methods and compare the results • Spend more time on data cleaning, preparation and transformation