Download as PDF, PPTX
![13
Naïve Bayes Learning
Naïve Bayes algorithm
Naïve Bayes classification
– Assumption that all input features are conditionally independent!
– MAP classification rule: for
)|()|()|(
)|,,()|(
)|,,(),,,|()|,,,(
21
21
22121
CXPCXPCXP
CXXPCXP
CXXPCXXXPCXXXP
n
n
nnn
Lnn ccccccPcxPcxPcPcxPcxP ,,,),()]|()|([)()]|()|([ 1
*
1
***
1
),,,( 21 nxxx x](https://image.slidesharecdn.com/fptis-churnprediction-150627095940-lva1-app6892/85/churn-prediction-in-telecom-13-320.jpg)
![13
Naïve Bayes Learning
Naïve Bayes algorithm
Naïve Bayes classification
– Assumption that all input features are conditionally independent!
– MAP classification rule: for
)|()|()|(
)|,,()|(
)|,,(),,,|()|,,,(
21
21
22121
CXPCXPCXP
CXXPCXP
CXXPCXXXPCXXXP
n
n
nnn
Lnn ccccccPcxPcxPcPcxPcxP ,,,),()]|()|([)()]|()|([ 1
*
1
***
1
),,,( 21 nxxx x](https://image.slidesharecdn.com/fptis-churnprediction-150627095940-lva1-app6892/75/churn-prediction-in-telecom-13-2048.jpg)
Uploaded byHong Bui Van
churn prediction in telecom
The document discusses machine learning algorithms for predicting customer churn in a prepaid mobile network. It presents an overview of supervised and unsupervised learning techniques including support vector machines, k-nearest neighbors, neural networks, decision trees and naive Bayes. The document outlines features for a churn prediction model, describes a demo of the model using different algorithms, and evaluates the classification accuracy and churn rates.
In this document
Powered by AI
Introduction of the presenter, Bui Van Hong, with contact information.
Agenda overview detailing the main topics: churn prediction, machine learning methods, models, features, and demo.
Introduces machine learning concepts: supervised, unsupervised, semi-supervised, reinforcement learning, and deep learning.
Defines customer churn in prepaid services, focusing on inferring lack of activity as a prediction challenge.
Illustrates the customer states: New, Active, Inactive, and Churned.
Details the data pipeline for churn prediction involving training and testing data and model learning algorithms.
Describes evaluation metrics for classification accuracy including TP, FP, TN, and FN rates.
Reiterates the agenda, emphasizing the machine learning algorithms and demo.
Introduction to Support Vector Machines, includes details about linear and non-linear functions.
Introduction to the k-Nearest Neighbor algorithm for churn prediction.
Explains the architecture and learning process of the Multilayer Perceptron (MLP) neural network.
Overview of decision tree algorithm focusing on classification of customers.
Explains the Naive Bayes classification method and its assumption of feature independence.
Continuation of the agenda focusing on logical design and demo.
Lists and explains the attributes used in the churn prediction model like outgoing calls and credit.
Describes the logical design framework, including input-output descriptions and algorithms used.
Provides details of the attributes in training data and the structure for churn classification.
Elaborates additional input data attributes used for churn prediction analysis.
Repeats the data sampling structure for input attributes used in churn prediction.
Shows the results of churn prediction algorithms with accuracy and churn rate comparison.
No content provided for Slide 21, implied closing of the presentation.
![[WMD 2016] SurveyMonkey >> Elena Verna "Minimize involuntary churn through p...](https://cdn.slidesharecdn.com/ss_thumbnails/involuntarychurn-elenaverna-161104035436-thumbnail.jpg?width=600ounds&width=560&fit=bounds)
![Deriving economic value for CSPs with Big Data [read-only]](https://cdn.slidesharecdn.com/ss_thumbnails/derivingeconomicvalueforcspswithbigdata-final11sep2013read-only-130911045922-phpapp02-thumbnail.jpg?width=600ounds&width=560&fit=bounds)
churn prediction in telecom
- 1. May, 2015 Bui VanHong Email: hongbv@fpt.com.vn
- 2. Agenda Churn prediction inprepaid mobile telecommunication network Machine Learning Introduction customer churn Diagram of possible customer states Churn prediction Model Classification accuracy Machine learning algorithm Support vector machine Nearest neighbour machine Multilayer percenptron neural network machine Decision tree machine Native bayes machine Feature presentation & Demo List of attributes Build Training & testing data Demo
- 3. 3 Machine Learning Supervised learning Supervisedlearning is the machine learning task of inferring a function from labeled training data. unsupervised learning Supervised learning is the machine learning task of inferring a function from unlabeled training data. Semi-supervised learning falls between unsupervised learning (without any labeled training data) and supervised learning (with completely labeled training data). Reinforcement learning Reinforcement learning is an area of machine learning inspired by behaviorist psychology (chess player) Deep learning Deep learning (deep machine learning, or deep structured learning, or hierarchical learning, or sometimes DL) is a branch of machine learning based on a set of algorithms that attempt to model high-level abstractions in data by using model architectures, with complex structures or otherwise, composed of multiple non-linear transformations
- 4. 4 Introduction - Customer churnis define as the lose of customers - Churn in prepaid service is actually measure base on the lack of activity in the network over a period time, thus there is no formal notification from customer of ending a contract term, our goal is to infer when this lack of activity may happen in the future for each active customer - Churn prediction can be viewed as supervised classification problem where the behavior of previously know churners an non-churners are used to train a binary classified
- 5. 5 Diagram of possiblecustomer states Each customer can be in one of the following states: New, active, inactive or churn
- 6. 6 Churn prediction Model Trainingdata Testing data RealModelFeature extraction Requires Expert Knowledge in Telco industry Learning algorithm & Prediction Independent (1) Build (2) Test (3) Test=OK (4) Predict (5) Update ETL 1 ETL 2
- 7. 7 Classification Accuracy Tiêu chíđánh giá độ chính xác của kết quả Actual class/Predictive class Churned Non-churned Churned True Positive (TP) False Negative (FN) Non-churned False Positive (FP) True Negative(TN) Accuracy= 𝑇𝑃+𝑇𝑁 𝑇𝑃+𝑇𝑁+𝐹𝑃+𝐹𝑁 Churn rate = 𝑇𝑃 𝑇𝑃+𝐹𝑃
- 8. Agenda Churn prediction inprepaid mobile telecommunication network Introduction Diagram of possible customer states Churn prediction Model Classification accuracy Machine learning algorithm Support vector machine Nearest neighbor machine Multilayer perceptron neural network machine Decision tree machine Native Bayes machine Feature presentation & Demo List of attributes Demo-POC
- 9. 9 Support Vector MachineLearning Non-linear/Linear SVM Linear Functions ( ) T g b x w x Nonlinear Functions =-1 =+1
- 10. 10 Nearest Neighbor Learning k-NearestNeighbor Nearest Neighbor =-1 =+1
- 11. 11 Multilayer Perceptron (MLP)Neural Network Learing MLP algorithm 1 2 3 x1 x2 =-1 =+1
- 12. 12 Decision Tree Learning DecisionTree algorithm Decision Tree =-1 =+1 Y:churn N: non-churn
- 13. 13 Naïve Bayes Learning NaïveBayes algorithm Naïve Bayes classification – Assumption that all input features are conditionally independent! – MAP classification rule: for )|()|()|( )|,,()|( )|,,(),,,|()|,,,( 21 21 22121 CXPCXPCXP CXXPCXP CXXPCXXXPCXXXP n n nnn Lnn ccccccPcxPcxPcPcxPcxP ,,,),()]|()|([)()]|()|([ 1 * 1 *** 1 ),,,( 21 nxxx x
- 14. Agenda Churn prediction inprepaid mobile telecommunication network Introduction Diagram of possible customer states Churn prediction Model Classification accuracy Machine learning algorithm Support vector machine Nearest neighbor machine Multilayer perceptron neural network machine Decision tree machine Native Bayes machine Feature presentation & Demo List of attributes Logical design Demo Evaluation & compartion
- 15. 15 List of attribute AttributeName Description Number Of Outgoing calls Number Of Outgoing calls Number Of Messages Sent Number Of Messages Sent Usage GPRS Subscriber is provided with GPRS service (Yes, No) Total MOU Minutes of usage ( MoU) per subscriber Total Number Of Recharges Total Number Of Recharges Total Credit Revenue Credit Revenue Per subscriber (Voice, SMS, GPRS, VAS) Total Bonus Revenue Bonus Revenue Per subscriber (Voice, SMS, GPRS, VAS) Customer Relationship Age classifies measures pertaining to Customers according to the number of years for which the Customer has had a relationship with the Telecommunications Services Provider. Remaining credit Remaining credit per subscriber Churn Churning customer status (yes, No)
- 16. 16 Logical Design Input Description DataBI system Feature Domain expert in industry Algorithm SVM, NB, Decision tree, MLP, k-Nearest Neighbour Output Description Churn/ no-churn Specify churn/no-churn for each subscribers
- 17. 17 Demo Sampling for trainingdata Attribute Description NUM_OG_CALLS Number of outgoing call for Month -1 SUM_DURATION_OG Minutes of usage ( MoU) per subscriber NUM_SMO Number Of Messages Sent for Month -1 NUM_DATA_UP Subscriber is provided with GPRS service (Yes, No) for Month -1 NUM_OG_CALLS_1 Number of outgoing call for Month -2 SUM_DURATION_OG_1 Minutes of usage ( MoU) per subscriber for Month -2 NUM_SMO_1 Number Of Messages Sent for Month -2 NUM_DATA_UP_1 Subscriber is provided with GPRS service (Yes, No) for Month -2 Recharge Number of rechard times for Month -1 Recharge_1 Number of rechard times for Month -2 RemainCredit Remain credit in Month -3 RemainCredit_1 Remain credit in Month -2 RemainCredit_2 Remain credit in Month -1 TOTAL_BONUS Revenue bonus of Month -2 TOTAL_CREDIT Revenue credit of Month -2 TOTAL_BONUS_1 Revenue bonus of Month -1 TOTAL_CREDIT_1 Revenue credit of Month -1 CLASS {1,0} Chun/no-churn Total records: 2876 rows
- 18. 18 Demo Sampling for Inputdata Attribute Description NUM_OG_CALLS Number of outgoing call for Month -1 SUM_DURATION_OG Minutes of usage ( MoU) per subscriber NUM_SMO Number Of Messages Sent for Month -1 NUM_DATA_UP Subscriber is provided with GPRS service (Yes, No) for Month -1 NUM_OG_CALLS_1 Number of outgoing call for Month -2 SUM_DURATION_OG_1 Minutes of usage ( MoU) per subscriber for Month -2 NUM_SMO_1 Number Of Messages Sent for Month -2 NUM_DATA_UP_1 Subscriber is provided with GPRS service (Yes, No) for Month -2 Recharge Number of rechard times for Month -1 Recharge_1 Number of rechard times for Month -2 RemainCredit Remain credit in Month -3 RemainCredit_1 Remain credit in Month -2 RemainCredit_2 Remain credit in Month -1 TOTAL_BONUS Revenue bonus of Month -2 TOTAL_CREDIT Revenue credit of Month -2 TOTAL_BONUS_1 Revenue bonus of Month -1 TOTAL_CREDIT_1 Revenue credit of Month -1 CLASS {1,0} ? Total records: 15175 rows
- 19. 19 Demo Sampling for Inputdata Attribute Description NUM_OG_CALLS Number of outgoing call for Month -1 SUM_DURATION_OG Minutes of usage ( MoU) per subscriber NUM_SMO Number Of Messages Sent for Month -1 NUM_DATA_UP Subscriber is provided with GPRS service (Yes, No) for Month -1 NUM_OG_CALLS_1 Number of outgoing call for Month -2 SUM_DURATION_OG_1 Minutes of usage ( MoU) per subscriber for Month -2 NUM_SMO_1 Number Of Messages Sent for Month -2 NUM_DATA_UP_1 Subscriber is provided with GPRS service (Yes, No) for Month -2 Recharge Number of rechard times for Month -1 Recharge_1 Number of rechard times for Month -2 RemainCredit Remain credit in Month -3 RemainCredit_1 Remain credit in Month -2 RemainCredit_2 Remain credit in Month -1 TOTAL_BONUS Revenue bonus of Month -2 TOTAL_CREDIT Revenue credit of Month -2 TOTAL_BONUS_1 Revenue bonus of Month -1 TOTAL_CREDIT_1 Revenue credit of Month -1 CLASS {1,0} ? Total records: 15175 rows
- 20. 20 Demo Result 0% 10% 20% 30% 40% 50% 60% 70% 80% Accuracy Churn rate Jan-15 NaiveBayes MultilayerPerceptronNearest neighbour Decision tree SVM Prediction actual- Churn actual-non- Churn Prediction actual- Churn actual- non-Churn Prediction actual- Churn actual-non- Churn Prediction actual- Churn actual- non-Churn Prediction actual- Churn actual- non- Churn Churn 7353 2955 4398 6550 2956 3594 5980 2906 3074 6137 2956 3181 3161 2497 664 non-Churn 7823 3596 4227 8626 3595 5031 9196 3645 5551 9039 3595 5444 12015 4054 7961 Accuracy 47% 53% 56% 55% 69% Churn rate 45% 45% 44% 45% 38%
- 21.