Hybrid Rule-Based and Machine Learning Framework for Embedding Anti-Discrimination Law in Automated Decision Systems

Abstract

Healthcare, predictive policing, content recommendation, and autonomous driving are all being impacted by AI, yet the findings it provides are frequently biassed. Representational equality can be undermined by symbolic connotations, and resource distribution can be affected by distributive implications. Predictive coding approaches have limitations due to static preprocessing and a single learning strategy, while fairness-aware machine learning helps to reduce bias and ensures that practitioners are held responsible. The majority of solutions, however, presume that discrimination is based on a single protected trait, such as gender or race. By incorporating anti-discrimination principles into automated decision-making systems, our Hybrid Rule-Based and Machine Learning Framework is able to overcome these limitations. They integrate robust predictive modelling with legal rule-based limitations using CNNs and LSTM networks. To account for both complicated, evolving legal standards and overlapping protected features, this hybrid architecture is employed. Experiments showed that the CNN-LSTM model resolved discriminatory law variability and AI decision-making unpredictability, and it achieved an accuracy of 97.07 %. By bridging the gap between legislative mandates and technology implementation, this method improves AI that is fair and makes decision-making more accountable in many circumstances.

Ähnliche Arbeiten