Predictive Models

DSA's Approach To Predictive Modeling

DSA does not use a “canned” approach to developing predictive models. Rather, the specific steps and methodologies employed are chosen according to modeling objectives, nature of the dependent variable(s), type and amount of data available and other client needs.

We support the full range of data mining tasks, including:

Classification
Estimation
Prediction
Affinity Grouping
Clustering
Description

We have applied these techniques to many different types of client objectives. Some examples include:

Attrition
New member performance projection
Brand affinity
Lead generation conversion
Brand switching
List segmentation
Cash flow
Product affinity
Cross selling
Profitability
Customer acquisition
Response analysis
Customer profiles
Segmentation / Customer clustering
Reactivation
2-Stage Gross/Net response

We employ a number of multivariate analytical techniques including:

Least Squares regression
Logistic regression
Bayesian Segment Models (random effects models)
Principal Components Analysis
Factor Analysis
Chi Square Automatic Interaction Detector
Cluster Analysis
Discriminant Analysis

Like many other companies we experiment with new predictive technologies such as Mixed Models, Genetic Algorithms, Hierarchical Bayes, etc. which show some promise for data mining applications. At this time, we do not advocate the use of Artificial Neural Networks (ANNs) because testing has shown that in its current form ANN appears to not outperform logistic regression.