This study developed a neural network-based multimodal fusion system for survival-risk prediction and auxiliary event-free classification. The system integrates histopathology feature vectors, RNA sequencing gene-expression data, and structured clinical variables using modality-specific encoders, attention-weighted fusion, a Cox survival-risk head, and an event-free classification head. The model was implemented in Python using PyTorch and evaluated using TCGA-BRCA, TCGALUAD, and a simulated local cohort. In TCGA-BRCA 5-fold cross-validation, the system achieved a mean concordance index (C-index) of 0.6573, classification accuracy of 0.8683, calibration error of 0.0271, and a worst single-missingmodality C-index drop of 0.0427. Genomic and histopathology modalities received the highest attention weights. However, transfer performance was weak on TCGA-LUAD and the simulated local cohort, with C-indices of 0.4682 and 0.4877, respectively. The findings support the model for breast cancer survival-risk prediction within TCGA-BRCA, while broader application requires cancer-specific retraining and validation using real Nigerian oncology data.

Redeemer Pyagbara Dobe, Chima Godknows Igiri , Onate Egerton Taylor, "Neural Network-Based Multimodal Fusion Model for Oncology Outcome Prediction: An Attention-Weighted Encoder-Fusion-Predictor Framework" International Journal of Computer Science and Engineering Development, V1(4): Page(59-67) September - October 2026. ISSN: 3139-0862. www.ijcsed.com. Published by Scientific and Academic Research Publishing.