Abstract
Although a strategy for predicting stock prices using relational data has been described recently, no practical way for selecting aggregating various forms of relational data to forecast stock prices has been discovered. The authors present an upgraded multilayer node graph attention network (FHAN) model that incorporates the Fraudar algorithm and provides insight into the interaction between several items. The model, which regards businesses as nodes and interactions as edges, aggregates data from various connection types and adds it to each company’s node representation, which is then automatically fed into the task-specific layer select information. The testing findings indicate that this approach is more accurate in stock price prediction than the currently popular neural network methodology. The experiment compares several distinct neural network algorithms. This approach is more accurate than the previous method when ideal parameters are used. The average rise is around 4%, while the largest increase is 24%.