|
CUSTOMER LOYALTY EVALUATION AND PREDICTION BASED ON DECISION TREE AND ARTIFICIAL NEURAL NETWORK: CASE OF OFOGH KOOROSH STORES IN TEHRAN |
|---|---|
| รหัสดีโอไอ | |
| Creator | Amirreza Estakhrian Haghighi, Abdolreza Shirazi |
| Title | CUSTOMER LOYALTY EVALUATION AND PREDICTION BASED ON DECISION TREE AND ARTIFICIAL NEURAL NETWORK: CASE OF OFOGH KOOROSH STORES IN TEHRAN |
| Contributor | - |
| Publisher | TuEngr Group |
| Publication Year | 2563 |
| Journal Title | International Transaction Journal of Engineering, Management, & Applied Sciences & Technologies |
| Journal Vol. | 11 |
| Journal No. | 5 |
| Page no. | 11A05O: 1-11 |
| Keyword | Decision Tree method, Artificial Neural Network (ANN), Learning algorithm, Customer loyalty prediction, Data Mining, Data training, K-Means clustering, Multilayer perceptron (MLP). |
| URL Website | http://TuEngr.com/Vol11_5.html |
| Website title | ITJEMAST V11(4) 2020 @ TuEngr.com |
| ISSN | 2228-9860 |
| Abstract | The secret to staying in the business world today is having satisfied and pleased customers who buy the company services over and over again and introduce such products/services to others. These companies need to know what the customer wants and how they can adapt to the customer's needs according to customer preferences over time. In this work, the factors affecting customer satisfaction and loyalty of Ofogh Koorosh stores were studied and then analyzed using data mining techniques and methods and the extent to which each factor influenced their loyalty. The results of the decision tree and artificial neural network (ANN) with different segmentation of observed data and evaluation criteria of the obtained models show that in the decision tree model with clustering criteria and dividing data set into six clusters the burden of computation and classification accuracy have been increased, and each criterion is initially prioritized within itself. The results of these clusters are combined and the results accurately predict customer loyalty. The proportionality between accuracy and readability criteria in this algorithm indicates that the considered criteria values are well averaged and have a uniform distribution because the detection rates of the samples with low priority over the samples compared with high priority are almost equal. In other words, criteria have well recognized the high and low-priority decision tree algorithm. In neural networks both in modeling (training) and in the validation (testing) it shows the high coefficient of explanation but with the observation of error rate of both modeling and validation it can be concluded that it is possible to have unbalanced data in our dataset. The error rate of the decision tree is less than the ANN according to F-values, indicating the decision tree as a whole is more successful in estimating burnout and prioritization than ANN. |
