Counterfactual Explanations and Federated Learning for Enhanced Data Analytics Optimisation

Breast cancer remains a critical problem. Its complexity requires robust methods to understand the underlying genetic interactions. This research applies a three-part approach: Federated Learning, Counterfactual Explanations and Structural Causal Models to analyse breast cancer gene expression data. First, we use the ability of Federated Learning to train on decentralised data samples, which allows us to gain deep insights into the different gene expressions in relation to the histological grades of breast cancer. To improve the interpretability of our model, counterfactual explanations were incorporated, allowing for a clearer ‘what-if’ analysis while revealing potential biases. Graph theory, used in conjunction with hierarchical Bayesian models, allowed us to visualise key gene interactions and shed light on their potential role in breast cancer pathogenesis. By applying structural causal models, we mapped a systematic landscape of causal interactions between genes that may play a critical role in cancer development. Using microarray datasets, our analysis has uncovered key genes and their significant correlations, laying the groundwork for the development of potential therapeutic interventions. Overall, this study paves the way for a holistic understanding of the genetics of breast cancer and offers promising approaches for treatment strategies.