You would need to perform hyper-parameter tuning of the model to get good results hopefully. I also believe it is vital to know very well the nature of your data (is it highly imbalanced?). If so, I would suggest using some oversampling techniques to see how it works, as we cannot simply tell without knowing the type and nature of the dataset. This paper Article Deep learning with knowledge transfer for explainable anomal...
here might help, as it utilises LSTMs for a real-world anomaly prediction application. Try out other similar models such as GRUs and CNNs as well as more conventional techniques (e.g. decision trees). Hope this helps,