Predictive Modeling is a process using known data to predict unknown consumer behavior. This process is made up of classification and/or estimation.
Classification: By examining features of a new object to assign the object to a predefined class.
Examples:
- Classifying credit applicants as low or high risk;
- Assigning customers to different segments by gender, age, income or purchasing behavior.
Estimation: Based on input data to estimate the unknown value of a continuous variable.
Examples:
- Estimate a customer's lifetime values
- Estimate a home values;
- Determine to whom a bank should offer a mortgage and how much.
Targeting via predictive modeling is becoming a common practice for marketing. But it is crucial to understand business objectives of each targeting project before modeling. The business objectives drive modeling techniques. Often several modeling techniques are applicable. Data miners need to select appropriate models and evaluate them properly. For example, to increase sales we model for acquisition, cross-sell and up-sell. We summarize the commonly used marketing strategies in the following table.
| Customer Acquisition |
Response Modeling |
| Acquire Profitable Customers |
Lifetime Value Modeling |
| Avoid High-Risk Customers |
Risk or Approval Modeling |
| Customer Knowledge: Attributes |
Profile Analysis |
| Customer Knowledge: Markets |
Cluster Analysis |
| Increase Customer's Value |
Up-Sell & Cross-Sell Modeling |
| Retain Profitable Customers |
Retention Modeling |
| Win Back Lost Customers |
Win-back Modeling |
| Improve Customer Satisfaction |
Market Research/Profiling |
Manifold's Predictive Modeling combines powerful dimension reduction techniques and regression techniques with non-parametric methods and provides the most reliable predictions. |
