This document summarizes a study on predicting customer churn for a major mobile provider. TreeNet models were used to predict the probability of customers churning (switching providers) within a 30-60 day period. TreeNet models significantly outperformed other methods, increasing accuracy and the proportion of high-risk customers identified. Applying the most accurate TreeNet models could translate to millions in additional annual revenue by helping the provider preemptively retain more customers.
Comparative Study of Machine Learning Algorithms for Sentiment Analysis with ...Sagar Deogirkar
Comparing the State-of-the-Art Deep Learning with Machine Learning algorithms performance on TF-IDF vector creation for Sentiment Analysis using Airline Tweeter Data Set.
Get to know in detail the termonologies of Random Forest with their types of algorithms used in the workflow along with their advantages and disadvantages of their predecessors.
Thanks, for your time, if you enjoyed this short article there are tons of topics in advanced analytics, data science, and machine learning available in my medium repo. https://medium.com/@bobrupakroy
ATTRIBUTE REDUCTION-BASED ENSEMBLE RULE CLASSIFIERS METHOD FOR DATASET CLASSI...csandit
Attribute reduction and classification task are an essential process in dealing with large data
sets that comprise numerous number of input attributes. There are many search methods and
classifiers that have been used to find the optimal number of attributes. The aim of this paper is
to find the optimal set of attributes and improve the classification accuracy by adopting
ensemble rule classifiers method. Research process involves 2 phases; finding the optimal set of
attributes and ensemble classifiers method for classification task. Results are in terms of
percentage of accuracy and number of selected attributes and rules generated. 6 datasets were
used for the experiment. The final output is an optimal set of attributes with ensemble rule
classifiers method. The experimental results conducted on public real dataset demonstrate that
the ensemble rule classifiers methods consistently show improve classification accuracy on the
selected dataset. Significant improvement in accuracy and optimal set of attribute selected is
achieved by adopting ensemble rule classifiers method.
Comparative Study of Machine Learning Algorithms for Sentiment Analysis with ...Sagar Deogirkar
Comparing the State-of-the-Art Deep Learning with Machine Learning algorithms performance on TF-IDF vector creation for Sentiment Analysis using Airline Tweeter Data Set.
Get to know in detail the termonologies of Random Forest with their types of algorithms used in the workflow along with their advantages and disadvantages of their predecessors.
Thanks, for your time, if you enjoyed this short article there are tons of topics in advanced analytics, data science, and machine learning available in my medium repo. https://medium.com/@bobrupakroy
ATTRIBUTE REDUCTION-BASED ENSEMBLE RULE CLASSIFIERS METHOD FOR DATASET CLASSI...csandit
Attribute reduction and classification task are an essential process in dealing with large data
sets that comprise numerous number of input attributes. There are many search methods and
classifiers that have been used to find the optimal number of attributes. The aim of this paper is
to find the optimal set of attributes and improve the classification accuracy by adopting
ensemble rule classifiers method. Research process involves 2 phases; finding the optimal set of
attributes and ensemble classifiers method for classification task. Results are in terms of
percentage of accuracy and number of selected attributes and rules generated. 6 datasets were
used for the experiment. The final output is an optimal set of attributes with ensemble rule
classifiers method. The experimental results conducted on public real dataset demonstrate that
the ensemble rule classifiers methods consistently show improve classification accuracy on the
selected dataset. Significant improvement in accuracy and optimal set of attribute selected is
achieved by adopting ensemble rule classifiers method.
Ijaems apr-2016-23 Study of Pruning Techniques to Predict Efficient Business ...INFOGAIN PUBLICATION
The shopping mall domain is a dynamic and unpredictable environment. Traditional techniques such as fundamental and technical analysis can provide investors with some tools for managing their shops and predicting their business growth. However, these techniques cannot discover all the possible relations between business growth and thus, there is a need for a different approach that will provide a deeper kind of analysis. Data mining can be used extensively in the shopping malls and help to increase business growth. Therefore, there is a need to find a perfect solution or an algorithm to work with this kind of environment. So we are going to study few methods of pruning with decision tree. Finally, we prove and make use of the Cost based pruning method to obtain an objective evaluation of the tendency to over prune or under prune observed in each method.
A Threshold fuzzy entropy based feature selection method applied in various b...IJMER
Large amount of data have been stored and manipulated using various database
technologies. Processing all the attributes for the particular means is the difficult task. To avoid such
difficulties, feature selection process is processed.In this paper,we are collect a eight various benchmark
datasets from UCI repository.Feature selection process is carried out using fuzzy entropy based
relevance measure algorithm and follows three selection strategies like Mean selection strategy,Half
selection strategy and Neural network for threshold selection strategy. After the features are selected,
they are evaluated using Radial Basis Function (RBF) network,Stacking,Bagging,AdaBoostM1 and Antminer
classification methodologies.The test results depicts that Neural network for threshold selection
strategy works well in selecting features and Ant-miner methodology works best in bringing out better
accuracy with selected feature than processing with original dataset.The obtained result of this
experiment shows that clearly the Ant-miner is superiority than other classifiers.Thus, this proposed Antminer
algorithm could be a more suitable method for producing good results with fewer features than
the original datasets.
Understanding the Applicability of Linear & Non-Linear Models Using a Case-Ba...ijaia
This paper uses a case based study – “product sales estimation” on real-time data to help us understand
the applicability of linear and non-linear models in machine learning and data mining. A systematic
approach has been used here to address the given problem statement of sales estimation for a particular set
of products in multiple categories by applying both linear and non-linear machine learning techniques on
a data set of selected features from the original data set. Feature selection is a process that reduces the
dimensionality of the data set by excluding those features which contribute minimal to the prediction of the
dependent variable. The next step in this process is training the model that is done using multiple
techniques from linear & non-linear domains, one of the best ones in their respective areas. Data Remodeling
has then been done to extract new features from the data set by changing the structure of the
dataset & the performance of the models is checked again. Data Remodeling often plays a very crucial and
important role in boosting classifier accuracies by changing the properties of the given dataset. We then try
to explore and analyze the various reasons due to which one model performs better than the other & hence
try and develop an understanding about the applicability of linear & non-linear machine learning models.
The target mentioned above being our primary goal, we also aim to find the classifier with the best possible
accuracy for product sales estimation in the given scenario.
Slide explaining the distinction between bagging and boosting while understanding the bias variance trade-off. Followed by some lesser known scope of supervised learning. understanding the effect of tree split metric in deciding feature importance. Then understanding the effect of threshold on classification accuracy. Additionally, how to adjust model threshold for classification in supervised learning.
Note: Limitation of Accuracy metric (baseline accuracy), alternative metrics, their use case and their advantage and limitations were briefly discussed.
Multiclass classification of imbalanced dataSaurabhWani6
Pydata Talk on Classification of imbalanced data.
It is an overview of concepts for better classification in imbalanced datasets.
Resampling techniques are introduced along with bagging and boosting methods.
Proposing an Appropriate Pattern for Car Detection by Using Intelligent Algor...Editor IJCATR
Nowadays, the automotive industry has attracted the attention of consumers, and product quality is considered as an
essential element in current competitive markets. Security and comfort are the main criteria and parameters of selecting a car.
Therefore, standard dataset of CAR involving six features and characteristics and 1728 instances have been used. In this paper, it
has been tried to select a car with the best characteristics by using intelligent algorithms (Random Forest, J48, SVM,
NaiveBayse) and combining these algorithms with aggregated classifiers such as Bagging and AdaBoostMI. In this study, speed
and accuracy of intelligent algorithms in identifying the best car have been taken into account.
Ijaems apr-2016-23 Study of Pruning Techniques to Predict Efficient Business ...INFOGAIN PUBLICATION
The shopping mall domain is a dynamic and unpredictable environment. Traditional techniques such as fundamental and technical analysis can provide investors with some tools for managing their shops and predicting their business growth. However, these techniques cannot discover all the possible relations between business growth and thus, there is a need for a different approach that will provide a deeper kind of analysis. Data mining can be used extensively in the shopping malls and help to increase business growth. Therefore, there is a need to find a perfect solution or an algorithm to work with this kind of environment. So we are going to study few methods of pruning with decision tree. Finally, we prove and make use of the Cost based pruning method to obtain an objective evaluation of the tendency to over prune or under prune observed in each method.
A Threshold fuzzy entropy based feature selection method applied in various b...IJMER
Large amount of data have been stored and manipulated using various database
technologies. Processing all the attributes for the particular means is the difficult task. To avoid such
difficulties, feature selection process is processed.In this paper,we are collect a eight various benchmark
datasets from UCI repository.Feature selection process is carried out using fuzzy entropy based
relevance measure algorithm and follows three selection strategies like Mean selection strategy,Half
selection strategy and Neural network for threshold selection strategy. After the features are selected,
they are evaluated using Radial Basis Function (RBF) network,Stacking,Bagging,AdaBoostM1 and Antminer
classification methodologies.The test results depicts that Neural network for threshold selection
strategy works well in selecting features and Ant-miner methodology works best in bringing out better
accuracy with selected feature than processing with original dataset.The obtained result of this
experiment shows that clearly the Ant-miner is superiority than other classifiers.Thus, this proposed Antminer
algorithm could be a more suitable method for producing good results with fewer features than
the original datasets.
Understanding the Applicability of Linear & Non-Linear Models Using a Case-Ba...ijaia
This paper uses a case based study – “product sales estimation” on real-time data to help us understand
the applicability of linear and non-linear models in machine learning and data mining. A systematic
approach has been used here to address the given problem statement of sales estimation for a particular set
of products in multiple categories by applying both linear and non-linear machine learning techniques on
a data set of selected features from the original data set. Feature selection is a process that reduces the
dimensionality of the data set by excluding those features which contribute minimal to the prediction of the
dependent variable. The next step in this process is training the model that is done using multiple
techniques from linear & non-linear domains, one of the best ones in their respective areas. Data Remodeling
has then been done to extract new features from the data set by changing the structure of the
dataset & the performance of the models is checked again. Data Remodeling often plays a very crucial and
important role in boosting classifier accuracies by changing the properties of the given dataset. We then try
to explore and analyze the various reasons due to which one model performs better than the other & hence
try and develop an understanding about the applicability of linear & non-linear machine learning models.
The target mentioned above being our primary goal, we also aim to find the classifier with the best possible
accuracy for product sales estimation in the given scenario.
Slide explaining the distinction between bagging and boosting while understanding the bias variance trade-off. Followed by some lesser known scope of supervised learning. understanding the effect of tree split metric in deciding feature importance. Then understanding the effect of threshold on classification accuracy. Additionally, how to adjust model threshold for classification in supervised learning.
Note: Limitation of Accuracy metric (baseline accuracy), alternative metrics, their use case and their advantage and limitations were briefly discussed.
Multiclass classification of imbalanced dataSaurabhWani6
Pydata Talk on Classification of imbalanced data.
It is an overview of concepts for better classification in imbalanced datasets.
Resampling techniques are introduced along with bagging and boosting methods.
Proposing an Appropriate Pattern for Car Detection by Using Intelligent Algor...Editor IJCATR
Nowadays, the automotive industry has attracted the attention of consumers, and product quality is considered as an
essential element in current competitive markets. Security and comfort are the main criteria and parameters of selecting a car.
Therefore, standard dataset of CAR involving six features and characteristics and 1728 instances have been used. In this paper, it
has been tried to select a car with the best characteristics by using intelligent algorithms (Random Forest, J48, SVM,
NaiveBayse) and combining these algorithms with aggregated classifiers such as Bagging and AdaBoostMI. In this study, speed
and accuracy of intelligent algorithms in identifying the best car have been taken into account.
Improve Your Regression with CART and RandomForestsSalford Systems
Why You Should Watch: Learn the fundamentals of tree-based machine learning algorithms and how to easily fine tune and improve your Random Forest regression models.
Abstract: In this webinar we'll introduce you to two tree-based machine learning algorithms, CART® decision trees and RandomForests®. We will discuss the advantages of tree based techniques including their ability to automatically handle variable selection, variable interactions, nonlinear relationships, outliers, and missing values. We'll explore the CART algorithm, bootstrap sampling, and the Random Forest algorithm (all with animations) and compare their predictive performance using a real world dataset.
Using CART For Beginners with A Teclo Example DatasetSalford Systems
Familiarize yourself with CART Decision Tree technology in this beginner's tutorial using a telecommunications example dataset from the 1990s. By the end of this tutorial you should feel comfortable using CART on your own with sample or real-world data.
Churn in the Telecommunications Industryskewdlogix
Strategic Business Analysis Capstone Project Telecommunications Churn Management
Churn is a significant problem that costs telecommunications companies billions of dollars through lost revenue. Now that the market is more mature, the only way for a company to grow is to take their competitors customers. This issue
combined with the greater choice that consumers have gained means that any adverse touch point with a consumer can result in a lost customer.
Machine learning session6(decision trees random forrest)Abhimanyu Dwivedi
Concepts include decision tree with its examples. Measures used for splitting in decision tree like gini index, entropy, information gain, pros and cons, validation. Basics of random forests with its example and uses.
"Multilayer perceptron (MLP) is a technique of feed
forward artificial neural network using back
propagation learning method to classify the target
variable used for supervised learning. It consists of multiple layers and non-linear activation allowing it to distinguish data that is not linearly separable."
Predict Backorder on a supply chain data for an OrganizationPiyush Srivastava
Performed cleaning and founded the important variables and created a best model using different classification techniques (Random Forest, Naïve Bayes, Decision tree, KNN, Neural Network, Support Vector Machine) to predict the back-order for an organization using the best modelling and technique approach.
Credit Card Fraudulent Transaction Detection Research PaperGarvit Burad
Credit Card Fraudulent Transaction Detection Research Paper using Machine Learning technologies like Logistic Regression, Random Forrest, Feature Engineering and various techniques to deal with highly skewed dataset
Prediction of customer propensity to churn - Telecom IndustryPranov Mishra
The aim of this project is to help a telecom company with insights on customer behavior that would be useful for retention of customers. The specific goals expected to be achieved are given below
1. Identification of the top variables driving likelihood of churn
2. Build a predictive model to identify customers who have highest probability to terminate services with the company.
3. Build a lift chart for optimization of efforts by targeting most of the potential churns with least contact efforts. Here with 30% of the total customer pool, the model accurately provides 33% of total potential churn candidates.
Models tried to arrive at the best are
1. Simple Models like Logistic Regression & Discriminant Analysis with different thresholds for classification
2. Random Forest after balancing the dataset using Synthetic Minority Oversampling Technique (SMOTE)
3. Ensemble of five individual models and predicting the output by averaging the individual output probabilities
4. Xgboost algorithm
THE IMPLICATION OF STATISTICAL ANALYSIS AND FEATURE ENGINEERING FOR MODEL BUI...ijcseit
Scrutiny for presage is the era of advance statistics where accuracy matter the most. Commensurate
between algorithms with statistical implementation provides better consequence in terms of accurate
prediction by using data sets. Prolific usage of algorithms lead towards the simplification of mathematical
models, which provide less manual calculations. Presage is the essence of data science and machine
learning requisitions that impart control over situations. Implementation of any dogmas require proper
feature extraction which helps in the proper model building that assist in precision. This paper is
predominantly based on different statistical analysis which includes correlation significance and proper
categorical data distribution using feature engineering technique that unravel accuracy of different models
of machine learning algorithms.
THE IMPLICATION OF STATISTICAL ANALYSIS AND FEATURE ENGINEERING FOR MODEL BUI...IJCSES Journal
Scrutiny for presage is the era of advance statistics where accuracy matter the most. Commensurate between algorithms with statistical implementation provides better consequence in terms of accurate prediction by using data sets. Prolific usage of algorithms lead towards the simplification of mathematical models, which provide less manual calculations. Presage is the essence of data science and machine learning requisitions that impart control over situations. Implementation of any dogmas require proper feature extraction which helps in the proper model building that assist in precision. This paper is predominantly based on different statistical analysis which includes correlation significance and proper categorical data distribution using feature engineering technique that unravel accuracy of different models of machine learning algorithms.
Data Science - Part V - Decision Trees & Random Forests Derek Kane
This lecture provides an overview of decision tree machine learning algorithms and random forest ensemble techniques. The practical example includes diagnosing Type II diabetes and evaluating customer churn in the telecommunication industry.
TreeNet Tree Ensembles & CART Decision Trees: A Winning CombinationSalford Systems
Understand CART decision tree pros/cons, how TreeNet stochastic gradient boosting ca n help overcome single-tree challenges, and what the advantages are when using CART and TreeNet in combination for predictive modeling success.
When building a predictive model in SPM, you'll want to know exactly what you did to get your results. This short slide deck will show you how to review your work in the session logs.
Neuro-symbolic is not enough, we need neuro-*semantic*Frank van Harmelen
Neuro-symbolic (NeSy) AI is on the rise. However, simply machine learning on just any symbolic structure is not sufficient to really harvest the gains of NeSy. These will only be gained when the symbolic structures have an actual semantics. I give an operational definition of semantics as “predictable inference”.
All of this illustrated with link prediction over knowledge graphs, but the argument is general.
Accelerate your Kubernetes clusters with Varnish CachingThijs Feryn
A presentation about the usage and availability of Varnish on Kubernetes. This talk explores the capabilities of Varnish caching and shows how to use the Varnish Helm chart to deploy it to Kubernetes.
This presentation was delivered at K8SUG Singapore. See https://feryn.eu/presentations/accelerate-your-kubernetes-clusters-with-varnish-caching-k8sug-singapore-28-2024 for more details.
UiPath Test Automation using UiPath Test Suite series, part 3DianaGray10
Welcome to UiPath Test Automation using UiPath Test Suite series part 3. In this session, we will cover desktop automation along with UI automation.
Topics covered:
UI automation Introduction,
UI automation Sample
Desktop automation flow
Pradeep Chinnala, Senior Consultant Automation Developer @WonderBotz and UiPath MVP
Deepak Rai, Automation Practice Lead, Boundaryless Group and UiPath MVP
Dev Dives: Train smarter, not harder – active learning and UiPath LLMs for do...UiPathCommunity
💥 Speed, accuracy, and scaling – discover the superpowers of GenAI in action with UiPath Document Understanding and Communications Mining™:
See how to accelerate model training and optimize model performance with active learning
Learn about the latest enhancements to out-of-the-box document processing – with little to no training required
Get an exclusive demo of the new family of UiPath LLMs – GenAI models specialized for processing different types of documents and messages
This is a hands-on session specifically designed for automation developers and AI enthusiasts seeking to enhance their knowledge in leveraging the latest intelligent document processing capabilities offered by UiPath.
Speakers:
👨🏫 Andras Palfi, Senior Product Manager, UiPath
👩🏫 Lenka Dulovicova, Product Program Manager, UiPath
Elevating Tactical DDD Patterns Through Object CalisthenicsDorra BARTAGUIZ
After immersing yourself in the blue book and its red counterpart, attending DDD-focused conferences, and applying tactical patterns, you're left with a crucial question: How do I ensure my design is effective? Tactical patterns within Domain-Driven Design (DDD) serve as guiding principles for creating clear and manageable domain models. However, achieving success with these patterns requires additional guidance. Interestingly, we've observed that a set of constraints initially designed for training purposes remarkably aligns with effective pattern implementation, offering a more ‘mechanical’ approach. Let's explore together how Object Calisthenics can elevate the design of your tactical DDD patterns, offering concrete help for those venturing into DDD for the first time!
GraphRAG is All You need? LLM & Knowledge GraphGuy Korland
Guy Korland, CEO and Co-founder of FalkorDB, will review two articles on the integration of language models with knowledge graphs.
1. Unifying Large Language Models and Knowledge Graphs: A Roadmap.
https://arxiv.org/abs/2306.08302
2. Microsoft Research's GraphRAG paper and a review paper on various uses of knowledge graphs:
https://www.microsoft.com/en-us/research/blog/graphrag-unlocking-llm-discovery-on-narrative-private-data/
Generating a custom Ruby SDK for your web service or Rails API using Smithyg2nightmarescribd
Have you ever wanted a Ruby client API to communicate with your web service? Smithy is a protocol-agnostic language for defining services and SDKs. Smithy Ruby is an implementation of Smithy that generates a Ruby SDK using a Smithy model. In this talk, we will explore Smithy and Smithy Ruby to learn how to generate custom feature-rich SDKs that can communicate with any web service, such as a Rails JSON API.
Smart TV Buyer Insights Survey 2024 by 91mobiles.pdf91mobiles
91mobiles recently conducted a Smart TV Buyer Insights Survey in which we asked over 3,000 respondents about the TV they own, aspects they look at on a new TV, and their TV buying preferences.
Slack (or Teams) Automation for Bonterra Impact Management (fka Social Soluti...Jeffrey Haguewood
Sidekick Solutions uses Bonterra Impact Management (fka Social Solutions Apricot) and automation solutions to integrate data for business workflows.
We believe integration and automation are essential to user experience and the promise of efficient work through technology. Automation is the critical ingredient to realizing that full vision. We develop integration products and services for Bonterra Case Management software to support the deployment of automations for a variety of use cases.
This video focuses on the notifications, alerts, and approval requests using Slack for Bonterra Impact Management. The solutions covered in this webinar can also be deployed for Microsoft Teams.
Interested in deploying notification automations for Bonterra Impact Management? Contact us at sales@sidekicksolutionsllc.com to discuss next steps.
GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...James Anderson
Effective Application Security in Software Delivery lifecycle using Deployment Firewall and DBOM
The modern software delivery process (or the CI/CD process) includes many tools, distributed teams, open-source code, and cloud platforms. Constant focus on speed to release software to market, along with the traditional slow and manual security checks has caused gaps in continuous security as an important piece in the software supply chain. Today organizations feel more susceptible to external and internal cyber threats due to the vast attack surface in their applications supply chain and the lack of end-to-end governance and risk management.
The software team must secure its software delivery process to avoid vulnerability and security breaches. This needs to be achieved with existing tool chains and without extensive rework of the delivery processes. This talk will present strategies and techniques for providing visibility into the true risk of the existing vulnerabilities, preventing the introduction of security issues in the software, resolving vulnerabilities in production environments quickly, and capturing the deployment bill of materials (DBOM).
Speakers:
Bob Boule
Robert Boule is a technology enthusiast with PASSION for technology and making things work along with a knack for helping others understand how things work. He comes with around 20 years of solution engineering experience in application security, software continuous delivery, and SaaS platforms. He is known for his dynamic presentations in CI/CD and application security integrated in software delivery lifecycle.
Gopinath Rebala
Gopinath Rebala is the CTO of OpsMx, where he has overall responsibility for the machine learning and data processing architectures for Secure Software Delivery. Gopi also has a strong connection with our customers, leading design and architecture for strategic implementations. Gopi is a frequent speaker and well-known leader in continuous delivery and integrating security into software delivery.
Transcript: Selling digital books in 2024: Insights from industry leaders - T...BookNet Canada
The publishing industry has been selling digital audiobooks and ebooks for over a decade and has found its groove. What’s changed? What has stayed the same? Where do we go from here? Join a group of leading sales peers from across the industry for a conversation about the lessons learned since the popularization of digital books, best practices, digital book supply chain management, and more.
Link to video recording: https://bnctechforum.ca/sessions/selling-digital-books-in-2024-insights-from-industry-leaders/
Presented by BookNet Canada on May 28, 2024, with support from the Department of Canadian Heritage.
LF Energy Webinar: Electrical Grid Modelling and Simulation Through PowSyBl -...DanBrown980551
Do you want to learn how to model and simulate an electrical network from scratch in under an hour?
Then welcome to this PowSyBl workshop, hosted by Rte, the French Transmission System Operator (TSO)!
During the webinar, you will discover the PowSyBl ecosystem as well as handle and study an electrical network through an interactive Python notebook.
PowSyBl is an open source project hosted by LF Energy, which offers a comprehensive set of features for electrical grid modelling and simulation. Among other advanced features, PowSyBl provides:
- A fully editable and extendable library for grid component modelling;
- Visualization tools to display your network;
- Grid simulation tools, such as power flows, security analyses (with or without remedial actions) and sensitivity analyses;
The framework is mostly written in Java, with a Python binding so that Python developers can access PowSyBl functionalities as well.
What you will learn during the webinar:
- For beginners: discover PowSyBl's functionalities through a quick general presentation and the notebook, without needing any expert coding skills;
- For advanced developers: master the skills to efficiently apply PowSyBl functionalities to your real-world scenarios.
LF Energy Webinar: Electrical Grid Modelling and Simulation Through PowSyBl -...
Churn Modeling-For-Mobile-Telecommunications
1. N. Scott Cardell, Mikhail Golovnya, Dan Steinberg
Salford Systems
http://www.salford-systems.com
June 2003
2. Churn, the loss of a customer to a competitor, is a
problem for any provider of a subscription service or
recurring purchasable
◦ Costs of customer acquisition and win-back can be high
◦ Best if churn can be prevented by preemptive action or selection
of customers less likely to churn
Churn is especially important to mobile phone service
providers given the ease with which a subscriber can
switch services
The NCR Teradata center for CRM at Duke identified churn
prediction as a modeling topic deserving serious study
A major mobile provider offered data for an international
modeling accuracy and targeted marketing competition
3. Data was provided for 100,000 customers with at least 6 months
of service history, stratified into a roughly equal number of
churners and non-churners
Objective was to predict probability of loss of a customer 30-60
days into the future
Historical information provided in the form of
◦ Type and price of handset and recency of change/upgrade
◦ Total revenue and recurring charges
◦ Call behavior: statistics describing completed calls, failed calls, voice
and data calls, call forwarding, customer care calls, directory info
◦ Statistics included mean and range for at least 3 months, last 6
months, and lifetime
◦ Demographic and geographical information, including familiar Acxiom
style variables and census-derived neighborhood summaries.
4. Competition defined a sharply-defined task: churn within
a specific window for existing customers of a minimum
duration
Challenge was defined in a way to avoid complications of
censoring that could require survival analysis models
Each customer history was already summarized
Data quality was good
Vast majority of analytical effort could be devoted to
development of an accurate predictive model of a binary
outcome
5. Data Set Measure TreeNet
Ensemble
Single
TreeNet
2nd Best Avg. Std
Current Top Decile
Lift
2.90 2.88 2.80 2.14
(.536)
Current Gini .409 .403 .370 .269
(.096)
Future Top Decile
Lift
3.01 2.99 2.74 2.09
(.585)
Future Gini .400 .403 .361 .261
(.098)
6. Single TreeNet model always better than 2nd best
entry in field
Ensemble of TreeNets slightly better 3 out of 4
times
Best entries substantially better than the average
In broad telecommunications markets the added
accuracy and lift of TreeNet models over
alternatives could easily translate into millions of
dollars of revenue per year
7. A modest amount of data preprocessing was undertaken
to repair and extend original data
Some missing values could be recorded to “0”
Select non-missing values were recorded to missing
Experiments with missing value handling were conducted,
including the addition of missing value indicators to the
data
◦ CART imputation
◦ “All missings together” strategies in decision trees
Missings in a separate node
Missings go with non-missing high values
Missings go with non-missing low values
8. TreeNet was key to winning the tournament
◦ Provided considerably greater accuracy and top decile lift than any
other modeling method we tried
A new technology, different than standard boosting,
developed by Stanford University Professor Jerome Friedman
Based on the CART® decision tree and thus inherits these
characteristics:
◦ Automatic feature selection
◦ Invariant with respect to order-preserving transforms of
predictors
◦ Immune to outliers
◦ Built-in methods for handling missing values
9. Based on optimizing an objective function
◦ e.g: Likelihood function or sum of squared errors
Objective function expressed in terms of a target
function of the data
◦ The target function is fit as a nonparametric function of
the data
◦ The fit optimizes the objective function
Large number of small decision trees used to
form the nonparametric estimate
10. Current implementation allows:
◦ Binary classification
◦ Multinomial classification
◦ Least-squares regression
◦ Least-absolute-deviation regression
◦ M-regression (Huber loss function)
Other objective functions are possible
11. (Insert equation)
The dependent variable, y, is coded (-1,+1)
The target function, F(x), is ½ the log-odds ratio
F is initialized to the log odds on the full training
data set
(Insert equation)
◦ Equivalent to fitting data to a constant.
12. Do not use all training data in any one iteration
◦ Randomly sample from training data (we used a 50%
sample)
Compute log-likelihood gradient for each
observation
◦ (insert equation)
Build a K-node tree to predict G(y,x)
◦ K=9 gave the best cross-validated results
◦ Important that trees be much smaller than the size of an
optimal single CART tree
13. Let (insert equations)
Update formula (insert equation)
Repeat until T trees grown
Select the value of m≤T that produces the
best fit to the test data
14. Compute Ymn, a single Newton-Raphson step for
Bmn
(insert equation)
Use only a small fraction, p of, Ymn(Bmn=Pymn)
Apply the update formula
(insert equation)
15. P is called the learning rate, T is the number of trees
grown
The product pT is the total learning
◦ Holding pT constant, smaller p usually improves model fit to test
data, but can require many trees
Reducing the learning rate tends to slowly increase the
optimal amount of total learning
Very low learning rates can require many trees
Our CHURN models used values of p from 0.01 to 0.001
We used total learning of between 6 and 30
16. All the models used to score the data for the entries used
9-node trees
Our final models used the following three combinations:
◦ (p=.001; T=6000; pT=6)
◦ (p=.005; T=2500; pT=12.5)
◦ (p=.01; T=3000; pT=30)
One entry was a single TreeNet model (p=.01; T=3000;
pT=30)
◦ In this range all models had almost identical results on test data
◦ The scores were highly correlated (r≥.97)
◦ Within this range, a higher pT was the most important factor
◦ For models with pT=6, the smaller the learning rate the better