Data Shared Lasso: A Novel Tool to Discover Uplift.

Area of Science:

• Statistics
• Machine Learning
• Data Science

Background:

• Supervised learning models often assume homogeneity across data groups.
• Existing methods may struggle with sparse variations in regression coefficients between groups.
• Identifying subgroups that respond differently to interventions is crucial in many applications.

Purpose of the Study:

• To develop a flexible supervised learning model accommodating sparse variations in regression coefficients across pre-specified groups.
• To bridge the gap between individual group models and a single model for all groups.
• To introduce novel concepts and demonstrate the utility of the model in uplift analysis.

Main Methods:

• A novel supervised learning model is proposed for grouped data with sparse coefficient variations.
• The algorithm is designed for high-dimensional data frameworks.
• The model's performance is evaluated using sentiment analysis and credit card promotion datasets.

Main Results:

• The model demonstrates efficacy and interpretability in sentiment analysis.
• Application to a credit card promotion dataset reveals significant subgroup uplift.
• Targeting a specific subgroup with a promotion resulted in a 15% increase in purchase proportion.

Conclusions:

• The proposed model offers a powerful approach for analyzing grouped data with heterogeneous responses.
• It provides valuable tools for uplift analysis, enabling precise identification of responsive subgroups.
• The method has practical implications for targeted marketing and intervention strategies.