Feature selection is a fundamental data preprocessing step that significantly influences the performance of machine learning (ML) models by reducing irrelevant features. Although recent advancements in ML techniques have achieved high predictive accuracy, many operate as black boxes lacking explainability and requiring manual efforts for hyperparameter tuning and feature selection. Additionally, conventional feature reduction methods often perform poorly on structurally complex data. This research seeks to address some of these limitations by improving Feature Weighted Self-Organising Map (FWSOM), an interpretable and low-compute model capable of autonomous learning. FWSOM hypothesis is extended to support both clustered and path-based data, enabling automatic identification of relevant features in a transparent and computationally efficient manner, thereby facilitating more trustworthy and scalable AI systems.