<p>Water quality is important in the maintenance of aquatic organisms in aquaculture environment. Particularly, the occurrence of low oxygen level or anoxia is very dangerous to fish survival and the stability of aquatic ecosystem. It is important to detect cases of anoxia early to avoid irreversible damage. Currently, IoT based machine leaning and deep learning models are shown promise in aquaculture water quality monitoring. However, the existing models tend to be constrained in the resulting poor ability to model time-dependent effects, fitting to non-stationary water quality processes, and poor generalization in dynamic, real-world settings. To overcome these issues, the proposed model integrates the Internet of Things (IoT) and deep learning-based hybrid Quasi-Recurrent Neural Network, 1D convolutional layer with Long Short-Term Memory (QR-LSTM). The proposed QR-LSTM system coherently merged with a system to monitor and classify anoxia and non-anoxia conditions inside an aquaculture system in real-time. The IoT configuration constantly gathers the key water quality parameters, namely pH, temperature, and dissolved oxygen, and a tool of anoxia indices with annotated dataset is used to supervised learners. The proposed QR-LSTM architecture captures complicated spatiotemporal relationships in watery quality data. This hybrid architecture has the benefit of combining the speed of feature extraction of QRNN with the strength of long-term dependency modelling of LSTM to guarantee efficiency and robustness. Comprehensive results were carried out based on a public and real-time datasets of various aquaculture sites. The proposed model shows an impressive classification accuracy of 99.52% compared to the current Attention-NN model that shows an accuracy of 98.12%. The excellent performances indicate the high adaptability, accuracy and reliability of the suggested QR-LSTM model in detecting anoxic conditions in a variety of settings.</p>