Development of GRU Deep Learning Model for Predicting Daily United States Dollar to Tanzanian Shilling Exchange Rate Using Comparable Time-Lags Inputs

To import goods and services into the country, Tanzania relies on foreign currencies, specifically the United States Dollar (USD). Failure to timely predict accurate USD to TZS exchange rates may result in several problems, including failure to import into the country critical services and goods timely manner, losses in foreign exchange markets and bad decisions in investments. To address these challenges, this study has developed a Gated Recurrent Unit (GRU) Deep Learning model to predict the next day’s USD to TZS exchange rate (output) using three different inputs (time-lags) of previous days' exchange rates. This study has also developed a Web User Interface (UI) which is integrated with the developed GRU Deep Learning model. The Web UI receives the previous days' exchange rates entered by a user as inputs, predicts the next day’s exchange rate (output) and displays it to the user. The findings reveal that, 5-days time-lag (input) is the optimal (best performing) time-lag with a Mean Absolute Percentage Error (MAPE) score of 0.11%, followed by 10 days time-lag with a MAPE score of 0.20%  and 15 days time-lag with a MAPE score of 1.12%, suggesting that the shorter the time-lag (input), the better the performance of the GRU model in predicting the next day’s USD to TZS exchange rate (output). Therefore, this study recommends that Artificial Intelligence (AI) researchers and software developers use an optimal 5-day time-lag input when predicting the USD to TZS exchange rate using previous days' exchange rates using the GRU Deep Learning model. This study’s major contributions include an operational GRU model and Web User Interface (UI) for allowing users to predict daily USD to TZS exchange rates and a pre-processed 12-year-long daily USD to TZS exchange rates dataset ready and suitable for usage in AI research and software development activities