ProF-DXP: A Proxy-Based Fuzzy Framework for Data-Driven and Explainable Prediction of Life Satisfaction from Financial Satisfaction
Аннотация
Traditional quantitative models in socio-economic research often struggle to capture the inherent vagueness of human sentiment, while modern machine learning approaches frequently suffer from limited interpretability. This study proposes and evaluates the Proxy-Based Fuzzy framework for Data-Driven and Explainable Prediction (ProF-DXP), an end-to-end framework for constructing an interpretable fuzzy inference system from survey data. A key innovation lies in the use of concrete categorical variables as empirical proxies to define fuzzy sets over abstract universes of discourse. Membership functions for both the input and output are empirically derived using Kernel Density Estimation (KDE) based on the distributions of the proxy variables. The inferential core of the system consists of IF–THEN rules that are data-driven and extracted through fuzzy association rule mining. Defuzzification is performed using the Mean of Maxima method to obtain crisp life satisfaction predictions. In this study, the framework models the relationship between household financial and life satisfaction using World Values Survey Wave 7 data. The proposed model is compared with Linear Regression, Decision Tree regression, and a hyperparameter-tuned Multi-Layer Perceptron (MLP), using the same numerical input and output variables. Results indicate that while the fuzzy model exhibits lower predictive accuracy, it offers process-level explainability, enabling a transparent, step-by-step audit of the inference mechanism. This study contributes a replicable, data-driven pipeline for constructing white-box predictive models in the social sciences and demonstrates that, in research on human well-being, interpretability and transparency constitute a substantive methodological advantage alongside predictive performance.