The available models that predict total ammonia nitrogen concentration ([TAN]) in anaerobic digestion (AD) overestimate or underestimate the inhibition effect of [TAN], making them impractical to be implemented in full-scale digesters. Therefore, this study was conducted to establish a general and simple model to predict [TAN] with a decent predictive power by implementing machine learning modeling approaches. The experiment was conducted on AD reactors treating fish waste with three different feeding strategies and digestion times. Of 324 data entries, 80% and 20% were used for model training and testing. Cross-validation was also performed to determine the flexibility level and estimate the accuracy of the test dataset. Nine individual models were trained independently on training dataset, and the best-tuned models were inputted into voting regressor (VR) algorithm. VR outperformed other trained models with RMSE_test = 0.15, R²_test = 0.98, and 4.76% of model error. The model's generality was also confirmed by high coefficient of variation of several features used for model establishment, representing the complexity of AD process. Therefore, this model can be implemented to predict [TAN] in other AD systems, especially in systems treating high protein contents. It will be beneficial to maintain stability and improve the efficiency of AD.