Dynamic decision making simulation with limited data via causal inference.

Area of Science:

• Machine Learning
• Causal Inference
• Decision Science

Background:

• Machine learning is widely used in decision-making.
• Interventions in decision-making tasks are challenging due to limited data and cascading causal effects.
• Traditional methods using static data are insufficient for post-intervention analysis.

Purpose of the Study:

• To propose a novel framework for simulating interventions in decision-making tasks.
• To address the limitations of static data analysis in the presence of interventions.
• To infer post-intervention data by uncovering causal relationships.

Main Methods:

• Developed a framework to simulate interventions by inferring post-intervention data.
• Identified and utilized causal relationships within datasets.
• Designed two inference methods: direct computation of weights and model-fitted weights.
• Applied the framework to PCOS Prediction and Law School Admissions scenarios.

Main Results:

• The proposed framework realistically simulates intervention processes.
• The framework provides more reliable outcomes for machine learning tasks compared to static methods.
• Experimental results validated the framework's effectiveness in diverse scenarios.

Conclusions:

• The framework offers a robust solution for analyzing interventions in decision-making with limited data.
• Uncovering causal relationships is key to accurate post-intervention data inference.
• This approach enhances the reliability of machine learning models in dynamic decision-making environments.