Satellites in the geostationary orbit are located in the outermost radiation belt area and are vulnerable to space weather conditions, such as high energetic electron interactions, which can cause damage to satellite components. It is important to predict electron flux in this orbit to reduce the negative impact of these phenomena. Prediction models have been developed using different approaches, such as empirical, probabilistic, physics-based, and machine learning models. One of the methods used is the Long Short-Term Memory (LSTM). LSTM process inputs in the form of multidimensional data points where one data point may have multiple parameters. This allows a large number of parameter combinations to be served as input. Previous studies have employed the integral value of electron flux observed by Geostationary Operational Environmental Satellite (GOES) satellites as the main LSTM input due to its autocorrelation characteristic, along with other parameters from various space weather phenomena such as solar wind and geomagnetic indices. However, more parameters can be derived from the GOES data and are yet to be employed. In this study, we use satellite position data and parameters derived from electron flux data (average, minimum, maximum, data sub-point, Day-of-Year, local time) as input parameters to determine the impact of each parameter on the performance of LSTM in predicting ≥ 2 MeV electron flux. We used data with 12-h and 6-h resolutions to predict the electron flux value 12 and 6 h ahead, respectively. The results show that the combination of average, minimum, maximum, and data sub-points as input parameters has a significant impact on increasing the prediction result. The results reached 0.845 prediction efficiency (PE) for 6-h resolution and 0.803 for 12-h resolution. We also used the logarithmic electron flux value instead of the actual value and got even higher scores with the average PE of 0.925 for 6-h resolution and 0.893 for 12-h resolution.