As the world demands sustainable protein sources the global food supply increasingly depends on aquaculture activities. Maintaining optimal water quality continues to pose ongoing difficulties. Fish health and production levels depend heavily on temperature control combined with proper pH values along with Dissolved Oxygen (DO) standards for optimal performance. Current water quality monitoring operations require long durations of manual work and delayed data analysis which reduces the speed of timely responses. The proposed research implements Internet of Things (IoT) technology along with ensemble Machine Learning (ML) model to improve DO and Water Quality Index (WQI) prediction accuracy. A series of IoT enabled sensors operated at 15-minute interval across multiple days gathered extensive dataset which measured temperature and pH levels and DO content. A comparison of five machine learning prediction methods which included Polynomial Regression and Support Vector Regression (SVR), Random Forest and Gradient Boosting and XGBoost is presented in this article. The proposed ensemble model achieved exceptional results because it generated \(R^2\) values of 0.84 to 0.86 for DO measurements and WQI calculations approaching 0.91. The model’s ability to detect non-linear patterns enables exact real-time predictions which enable quick proactive decisions that minimize fish mortality rates. Advanced machine learning when combined with IoT technology enables improved operational efficiency along with sustainability enhancements. The proposed methodology optimizes water management systems for Catla and Tilapia fish farmers to enhance yield production while promoting an environmentally sustainable aquaculture industry.

Internet of Things (IoT) technologies in environmental monitoring provide real-time remote data acquisition capabilities and decision-making for different domains, including aquaculture. This study investigates a secure water quality monitoring system designed specifically for fish farms through IoT technology with an emphasis on lightweight encryption methods combined with real-time GSM communications. The primary objective is to protect and maintain the confidentiality, integrity, and operational efficiency of vital pH, Dissolved Oxygen (DO), and Temperature (T) measurements collected by low-power microcontroller devices. A wireless data transmission system was based on Arduino with calibrated analog sensors and the SIM800L GSM module to implement the MQTT protocol for data transfer. The data security assessment involved the implementation and testing of the AES-128-CBC and SPECK-128-CBC symmetric encryption algorithms under equivalent real-time circumstances. SPECK-128-CBC delivered faster encryption (209.64 µs) than AES-128-CBC (235.08 µs) along with improved memory efficiency and more compact payloads, indicating its best fit for constrained environments. AES-128-CBC demonstrated a slightly higher level of ciphertext entropy, reaching 5.18 bits/byte compared to 5.09 of SPECK-128-CBC. Using a weighted scoring method that weighted entropy at 40%, processing speed at 30%, and memory and payload efficiency at 30%, SPECK achieved 0.9931 while AES reached 0.9484. These results suggest that SPECK-128-CBC offers an energy-efficient encryption solution to provide optimal security for GSM-based IoT systems in aquaculture environments.