This preprint presents a hybrid neuro-symbolic framework for complex and composite Human Activity Recognition (HAR) in multi-user wearable sensing environments. Existing HAR systems primarily focus on single-user activity recognition and often struggle to model interaction dynamics, concurrent activities, and hierarchical activity compositions. To address these limitations, the proposed framework integrates multi-scale CNN-BiLSTM temporal feature extraction, Graph Neural Network (GNN)-based interaction modeling, attention-driven sensor fusion, and probabilistic symbolic reasoning within a unified architecture. The framework is evaluated using a newly developed multi-user wearable HAR dataset comprising synchronized wrist, waist, and chest-mounted IMU sensors collected from 30 participants across 200 recording sessions. The dataset includes atomic activities, interaction events, and composite collaborative tasks, enabling evaluation under realistic multi-user conditions. Experimental results demonstrate that the proposed approach achieves superior performance compared with classical machine learning, deep learning, Transformer-based, and symbolic-only baselines. Additional analyses include ablation studies, robustness evaluation, statistical significance testing, scalability assessment, deployment benchmarking, and comparison against the PAMAP2 benchmark dataset. To support reproducible research, this repository provides the manuscript, dataset documentation, source code, preprocessing pipelines, baseline implementations, evaluation scripts, and supplementary materials. The work contributes to the advancement of wearable sensing, neuro-symbolic artificial intelligence, graph-based activity modeling, and next-generation human activity recognition systems. Keywords: Human Activity Recognition, Wearable Sensors, Multi-User Activity Recognition, Neuro-Symbolic AI, Graph Neural Networks, Sensor Fusion, Deep Learning, Temporal Reasoning, IMU Sensors, Ambient Intelligence.