Framework Predictive Modeling

Problem Statement 

  • Business Context and Portfolio Performance Analysis
  • Understanding  Customers in case of Cross Sell or Response Modeling
  • Define Problem Statement e.g. Increase cross sell rate for Direct Mailer campaigns
  • Defining Modeling Structure


      • Target Variable
      • Performance Window
      • Observation Window
      • Exclusion Criteria

Data Preparation and Treatment

    • Create data for independent variables
    • Exploratory Data Analysis (EDA) Univariate analysis & Profiling
    • Outlier & missing value treatment
    • Variable transformations  and creating derived variables
    • Create Development and Validation data samples
    • Bivariate Analysis

Statistical Technique Selection

    • Depending on target variable type and measurement level relevant technique can be selected
    • Development time, business urgency, business context and deployment environment are other important considerations
    • For Binary Dependent: Logistic Regression, and Decision Tree can be considered

Model Build and Validation

    • Model can be build with available list of independent attribute sets
    • Model Performance criteria can be defined to accept the final model
    • Selected Model should be validated on Validation sample

Documentation and Deployment

    • Final validated model can be used for business application – cross sell model in campaigns
    • Model should be documented with all the assumptions  for monitoring performance on a regular basis
    • Analyst should support model deployment  and validation of the deployment

Predictive Model Applications across industry verticals