Forecasting currency prices presents a formidable challenge due to the inherent volatility of financial markets. This research introduces a novel model for predicting future currency market prices, employing Gated Recurrent Units (GRUs) neural networks. This research uses GRUs, which unlike traditional RNNs and their struggles with the vanishing gradient challenge, achieve better performance. It is a good approach because it improves the efficiency of the training. The internal designs of GRUs have been modified to tackle several problems such as avoiding getting stuck in local minima, decreasing the total time complexity, and overcoming problems encountered with stochastic gradient descent. These adjustments increase the effectiveness and reliability of the model's performance. The proposed model's efficiency is measured through root mean square error calculations on diverse datasets. Extensive experimentation conducted on a variety of datasets, shows that our proposed method is highly able to make successful and very accurate predictions of future currency prices.