An artificial neural network (ANN) model was developed to predict the bioremediation of POME using black fungus (Aspergillus niger) immobilised on coconut husk. A multilayer feedforward neural network (MFNN) model was developed using MATLAB software. The performance of 12 backpropagation training algorithms from 6 different classes (gradient descent with momentum, self-adaptive learning rate, conjugate gradient backpropagation, resilient backpropagation, Bayesian regulation propagation, and quasi-newton) were compared and analysed. Datasets of experiment were extracted and divided into training, validation, and testing data of the training algorithms. The input data were temperature, agitation speed, fermentation duration and POME concentration and output data were Chemical Oxygen Demand (COD) and turbidity reduction percentage. The best performing training algorithm was determined based on the consistency of the output prediction model errors namely, mean squared error (MSE), root mean squared error (RMSE), mean absolute percentage error (MAPE), and mean absolute error (MAE). The study finds that Bayesian Regulation Propagation (trainbr) is the most suitable training algorithm for reducing both COD and turbidity of POME. The trainbr resulted in the lowest error values of 1.173 (RMSE), 0.838 (MAPE), and 0.002 (MAE) for COD reduction, and 0.383 (RMSE), 0.238 (MAPE), and 0.002 (MAE) for turbidity reduction. Comparison between actual and predicted outputs show that ANN model with structure 4-1-15-2 trained using trainbr algorithm shows best performance in predicting the bioremediation of POME by immobilised black fungus with lowest MSE of 0.009 at epoch 205.