Monitoring water quality metrics of ponds with IoT sensors and machine learning to predict fish species survival

Abstract

Aquaculture involves cultivating various marine and freshwater aquatic creatures within regulated environments. Monitoring the aquatic environmental conditions in real-time is crucial for successful fish farming. The Internet of Things (IoT) offers significant potential for real-time monitoring, and this paper introduces an IoT framework designed for efficient monitoring and effective control of various water-related aquatic environmental parameters. The proposed system is implemented as an embedded system utilizing sensors and an Arduino microcontroller. In cultivating pond water, diverse sensors such as pH, temperature, and turbidity sensors are deployed, with each sensor connected to an Arduino Uno-based microcontroller board. These sensors collect data from the water, which is then stored as a CSV file in an IoT cloud platform called ThingSpeak through the Arduino microcontroller. To gather data for analysis, we conducted measurements across five ponds, varying in size and environmental conditions. After getting the real-time data, we compared our experimental results with the standard reference values. As a result, we could take the decision of whether a pond is suitable for cultivating fish or not. After that, we labeled the data with 11 fish categories: Katla, sing, prawn, shrimp, rui, tilapia, pangas, karpio, magur, silver carp, and koi. The data was analyzed using 10 machine learning (ML) algorithms, including J48, Random Forest, K Nearest Neighbors (K-NN), K*, Logistic Model Tree (LMT), Reduced Error Pruning Tree (REPTree), Jumping Rule Inference with Pruned Search (JRIP), Partial Decision Trees (PART), Decision Table, and Logit boost. After experimental analyses, it was discovered that only three of the five ponds were ideal for fish farming, and those three ponds only met the required standards for pH, Temperature, Turbidity, and Conductivity. Among the state-of-art machine learning algorithms, Random Forest achieved the highest score of performance metrics as accuracy 94.42%, kappa statistics 93.5%, and Avg. TP Rate 94.4%. In addition, we calculated the Biochemical Oxygen Demand (BOD), Chemical Oxygen Demand (COD), and Dissolved Oxygen (DO) for one scenario. This study includes prototype hardware details of the proposed IoT system.