The goal is to estimate the injector-to-producer connectivity from injection-production history data by implementing an attention-based graph neural network for fusion model (AGFM). The AGFM can identify the complex relationships between the injectors and producers, ensuring the spatially dense estimated injector-to-producer connectivity. The model is trained and tested on a dataset containing two types of injecting fluids: carbon dioxide (CO2) and water. The AGFM model correlates the relationships between every injector and all producers concerning produced oil, water, and gas. AGFM involves constructing a graph where nodes represent wells and edges represent their spatial and operational relationships. Node features include injection and production rates, while edge weights are based on physical proximity and connectivity among wells. The model can also optimize the water alternating gas ratio (WAG ratio) by alternating the injection of CO2 and water. The model can benefit from optimizing the WAG parameters to improve sweep efficiency and reduce gas channeling. We assessed the AGFM model through three scenarios of experiments. The first scenario uses CO2 as the injection fluid, the second uses water as the injection fluid, and the third uses CO2 and water alternately. We aligned the actual produced gas, oil, and water in each scenario for the methods with the model's predictions. We also compared the results with some selected state-of-the-art in terms of accuracy and mean squared error (MSE) (more quantitative and qualitative can be presented). The significant finding of the AGFM model was its ability to identify long-range dependencies between the injector and producer wells correctly. The AGFM model also correctly identified the immediate connections between each injector and all producer wells. Overall, the observational results support the conclusion that the AGFM model is a promising new approach for injector-producer connectivity estimation.