Predictive Modeling for Customer
Intelligence: The KDD Process Model
A Refresher on Data Preprocessing and Data Mining
Advanced Sampling Schemes
• cross-validation
(stratified, leave-one-out)
• bootstrapping
Neural networks
• multilayer perceptrons (MLPs)
• MLP types (RBF, recurrent,
etc.)
• weight learning (backpropagation,
conjugate gradient, etc.)
• overfitting, early stopping, and weight regularization
• architecture
selection (grid search, SNC, etc.)
• input selection
(Hinton graphs, likelihood statistics, brute force, etc.)
• self organizing maps (SOMs) for unsupervised
learning
• case
study: SOMs for country corruption analysis
Support Vector Machines (SVMs)
• linear programming
• the kernel trick and Mercer theorem
• SVMs for classification
and regression
• multiclass SVMs (one versus one, one versus
all coding)
• hyperparameter tuning using cross-validation methods
• case
study: benchmarking SVM classifiers
Opening up the Neural Network and SVM Black
Box
• rule extraction methods (pedagogical vs
decompositional approaches such as neurorule, neurolinear,
trepan, etc.)
• two-stage
models
A Recap of Decision Trees (C4.5, CART, CHAID)
Regression Trees
• splitting/stopping/assignment criteria
Ensemble Methods
• bagging, boosting, stacking, random forests
Alternative Rule Representation Formats
• rule types (oblique, M-of-N, fuzzy, etc.)
• decision tables
(lexicographical ordering, contraction methods, etc.)
• decision diagrams
• case study: decision tables and diagrams
for customer scoring
Bayesian Network Classifiers
• naive Bayes
• tree augmented naive Bayes (TAN)
• unrestricted Bayesian network
classifiers
• Bayesian inference
• case study: Bayesian networks for churn
prediction
Survival Analysis
• censoring
• Kaplan-Meier analysis
• parametric survival analysis
• proportional hazards regression
• neural networks for survival
analysis
• case study: neural network survival analysis for
customer scoring
Learning Using Networked Data
• Markov random fields
• homophily (guilt by association)
• local classifiers
• relational classifiers (relational neighbor,
probabilistic relational neighbor, relational logistic regression)
• collective
inference (Gibbs sampling, iterative classification, etc.)
Monitoring and Backtesting Analytical Models
• quantitative versus qualitative model monitoring
• model
backtesting (model stability, binomial/Hosmer-Lemeshow test,
traffic light indicator approach, impact of macro-economic
effects)
• model benchmarking (internal versus external benchmarking,
benchmarking statistics)
• qualitative validation of analytical
models (data quality, model design, documentation, involvement
of management)
• case study: backtesting a customer scoring
model
Other Predictive Modeling Techniques (Short)
• semi-supervised learning
• genetic algorithms
• fuzzy techniques
• ant colony optimization
• case study: Antminer+ |
