This study develops a model to forecast crossmarket linkage strengths for global stock market networks using dynamic edge weight graphs and optimized BiLSTM approach. It uses forty-three global market indices data of 14 years duration. The models are trained using historical network structure data over a series of time steps. Additionally, hyper-parameter tuning is performed to optimize performance. The model is validated using nested cross validation and are benchmarked against XGBoost and Deep Neural Network. Optimized models are experimented using multiple sets of network structures (unfiltered, filtered, low edge weighted networks) and optimum training sizes are obtained. Model trained with unfiltered networks demonstrated better relative predictability.