Free space optical communication (FSOC) has gained importance during the last few decades, among others, due to its higher data rate, which can go up to 2.5 Gbps1 in commercial systems, and its secure transmission. There are several challenges an FSOC channel encounters, one of which is atmospheric turbulence. Atmospheric turbulence can degrade the optical signal due to effects such as intensity scintillation and beam wandering. In this work, a machine learning algorithm has been optimized to forecast the scintillation index for the next 512 time steps. Meteorological data, such as air temperature, relative humidity, and wind speed, is obtained together with the scintillation index during an experiment along a 7 km propagation path in Dayton, OH. The data is divided into four parts corresponding to the four seasons and the data in each season is divided into training and validation data. Bi-directional long short-term memory (Bi-LSTM) models have been optimized and tuned to forecast the scintillation index. The mean squared error (MSE) is used to compare the predicted scintillation index with the measured scintillation index, and the adaptive moment estimation (Adam) optimizer is used to update the trainable parameters to minimize the MSE. The root mean squared error (RMSE) is used to validate the model predictions in the validation data. The training process is performed with different Bi-LSTM models on the training data for each season and the performance of the model is measured using the validation data for the corresponding season. The Bi-LSTM model predicts the scintillation index with a weighted average of the RMSE around 0.03 for all seasons.