Kiran babu T S, Manoj Challa

Turkish Online Journal of Qualitative Inquiry (TOJQI)

Volume 12, Issue 6, June 2021:321-328

Research Article

Prediction of Water Quality System for Aquaculture using Machine Learning

Kiran babu T S1, Manoj Challa2

Abstract
Indian aquaculture and fishing is one of the key stages of food production that provides
healthy food security which also provides employment, livelihoods and contribute to the
growth of Indian agriculture. Like all other aquatic organisms they also have a wide range of
tolerance, so fish farming of any species requires certain conditions that must be met. Water
quality is a critical factor in the processing of aquatic organisms. In this paper, we created an
application using ML to predict water boundaries and monitor fish ponds. The water level is
predicted in an hourly manner to ensure the growth and survival of aquatic life. The web
application is built using Flask to alert the user to critical situations. The impact of water
parameter changes can be effectively treated if the information is analyzed and water quality
is expected ahead of time.

Keywords: Aquaculture, Machine Learning, LSTM, Square Error

1
Department of CSE, CMR Institute of Technology, Bengaluru India, [email protected]
2
Department of CSE, CMR Institute of Technology, Bengaluru India, [email protected]
321
 Prediction of Water Quality System for Aquaculture using Machine Learning

Introduction

Studies have shown that over the past three decades the use of fish has been widely used.
This study serves as a contribution to increasing and stabilizing fish production. Fish farming
means a growing variety of marine species such as fishing fish, sport fish, ornamental fish,
mussels, mussels, crustaceans, seaweed, algae, and fish eggs to harvest, breed, and raise in
various places such as rivers, lakes, sea, and lakes.
There are various restrictions on marine life. One parameter depends on the other parameter
and indirectly affects aquatic organisms. In fish farming, acid and alkaline water should be
kept between 6 and 8. If the pH value is below 4.5, the fish will die. Aqua farmers rely on
self-assessment by knowing the limits of water, time-consuming and inaccurate as boundaries
can change over time. To overcome this problem, modern technology must be introduced into
agriculture. Therefore, it increases productivity and reduces losses by constantly monitoring
water quality parameters. The water parameters that need to be continuously checked are
Temperature, Melted Oxygen, PH, Turbidity, Salt, etc.
Our main goal is to increase the productivity of marine fish and sustainably save their lives.
To accurately predict water quality, ML Algorithms such as LSTM and ARIMA models are
analyzed and used as the most appropriate. In this project water boundary prediction is done
on an hourly basis. e.g., predict next hour values based on previous data. When the predicted
water quality appears to exceed the critical conditions it is immediately sent to the user to
take the necessary precautionary measures. The web application is built using Flask to
display alerts to users and data detection of various targeted water restrictions.
This paper focuses on the importance of water quality for aquaculture. It helps aqua farmers
to produce high-quality fish, which in turn helps the economy of the agricultural sector.
Machine learning helps to provide better, more accurate, and faster forecasts of water quality
based on previously collected data. Predictability analysis can help catch up relationships
between multiple variables that can help assess risk with a specific set of conditions. LSTM
NN is regarded as the best and provides very clear information for detecting and monitoring
water quality.

Related Work

Juntao Liu et al [1] are using the Simple Recurrent Unit (SRU) to create an accurate and
automated system for predicting cultural water quality. It focuses on predicting water quality
based on pH and temperature. The advanced progression method is used where the abstract
data is removed using a method called a fixed limit. Finally, it is presented with the SRU
model and compared with the RNN model which shows that SRU has high accuracy in
simultaneous difficulty.
A prediction model based on Smooth Support Vector Machine (SSVM) was proposed by
Wijayanti Nurul Khotimah [2] to predict water quality in water. The RMSE value obtained is
0.0275 indicating that SSVM is an effective means of predicting water quality in water.
J. Wang et al. [3] investigates the characteristics of strong interactions with the correction of
water quality parameter information and the disappearance of gradient and gradient
322
 Kiran babu T S, Manoj Challa

explosions caused by data training of the traditional RNN network model, etc. The structure
is shown on this page.
Dong Yao, Lei Cheng, QiuXuan Wu, Gong Zhang, Bei Wu, YuQing He [4] investigates how
to analyze and predict the quality of fishing using an electrochemical sensor array such as
melted oxygen, pH ammonia, and nitrogen carried by an unauthorized air vehicle.
(UAV) proposed. A floating T-shaped floating structure of a UAV can see the rise and fall of
water. The sensors in the structure enable real-time detection in the floating UAV system.
Sensor data is transmitted backstage via the wireless network. Thereafter a series of time
analyses are used to evaluate and predict water quality data. Finally, the conclusion is shown
in the application process. The results show that the system can assess the quality of fishing
water efficiently and has high accuracy (95.1%).

Proposed System

The proposed system architecture shows the complete working of the system which can be
divided into three main stages i.e. Data Pre-processing, Model development, and prediction
of results.The Figure 1 show the proposed system architecture for water quality predicition
for aquaculture.

Figure:1 Proposed System Architecture

Data Preprocessing:
In this stage first, Data cleaning has been done, meaning data is filtered and altered such that
unwanted data will be parted out. Then the filtered data is analyzed and rifted into training
and testing data sets. After that data is normalized to bring the values of all variables to the
same scale without deforming differences in the ranges of values.

323
 Prediction of Water Quality System for Aquaculture using Machine Learning

Model Development Process:

LSTM(Long Short Term Memory) model is used to design this prediction system.
Development of the system has occurred in two phases. The 1st phase model is built using
the training data set. In the 2nd phase, the model is evaluated to assess the performance and
accuracy of the system.
Prediction of Result:
In this stage, the model is run for many different unseen sets of values to assess the likely
future behavior of the system. And the result is shown on a web application for easy
understanding.

Implementatıon

Various tools and modules of python are used to create the proposed model. Jupyter
Notebook was used to work with our project. Many libraries of Python are used to create our
water quality prediction model and produce some visualizations for the same such as Pandas,
Numpy, Keras, Scikit-learn. Flask framework is used to build the web application. The
various modules involved in the project are:
1) Data Cleaning and Preparation
The dataset is stored in a data frame. The rows with NaN values are dropped from the data
frame. The data has been resampled and the daily average has been calculated. Then, the
resampled data is saved in another CSV file. The resultant dataset is used for the model
training.
2) Data Pre-processing
The dataset has been split into training and testing datasets. The training set is 80%, whereas
the testing set is 20% of the original dataset. MinMax normalization is used to transform the
values of both training and testing datasets in the range of -1 and 1 as the ranges of the
parameters are different.
3) Creating the model
To create the LSTM model, first, we create a Sequential model. Next, add a visible layer
with6 input, a hidden layer with 16 LSTM neurons, and an output layer that makes a four
values prediction. ReLU activation function is used for the LSTM neurons to help in
accomplishing better execution, helps to overcome the vanishing gradient problem, enabling
models to adapt faster and perform better.
4) Compiling the model
The model is compiled using the loss function and the optimizer before training the model.
Mean Squared Error is used as the loss function and Adam as the optimizer which is
anoptimization technique that can be utilized instead of the classical stochastic gradient
process to update network weights repetitively based on our training data.

324
 Kiran babu T S, Manoj Challa

5) Training the model

The model is trained with 100 epochs or iterations. A batch size of 1 has been used while
training. EarlyStopping callback is used so that the model can terminate itself when the
monitored feature has stopped improving.
6) Model Evaluation
MSE, RMSE, and R2 score between the actual values and the predicted values are calculated
for evaluation and these values are the best for the LSTM model than other models for our
application. To improve the accuracy, the LSTM model can be re-trained.

Figure: 2 Flowchart of the Water Quality Prediction Model

325
 Prediction of Water Quality System for Aquaculture using Machine Learning

Figure: 3 Flowchart of Prediction Subroutine

7) Prediction
The LSTM model is used to predict the values of water parameters such as pH, Dissolved
Oxygen, Temperature, and Turbidity.
The below flow charts (Figure 2 and Figure 3) show the step-by-step execution
implemented at the backend and frontend of the system.

Results And Dıscussıon

Discussion
An application is developed were the parameter values can be entered to the system. The
Dataset is collected from kaggle.com for predicition water for an Aquaculture.
Once the values of parameters are entered, the predicted values for the next hour are shown to
the user using the backend algorithm for prediction and an appropriate message is displayed.
When all the predicted parameters go out of range alert message is shown, When the
visualization button is clicked, the graphs for the various parameters are shown as in Figure
4, which can help the user understand the changes in parameters over time.

326
 Kiran babu T S, Manoj Challa

Figure:4 Visualisation of Parameter

Conclusıon And Future Scope

Water quality is an essential factor for aquaculture. In this project LSTM model is developed
to predict the value of pH, Temperature, Turbidity, Dissolved Oxygen which are the main
parameters for water quality in determining the aquatic species habitations.
The dataset fed into the model is the recorded information of water quality parameters which
are obtained from the USGS website which monitors the stations and maintains the database
and continuous values can be generated by the USGS REST service URL tool.
The datasets show recorded values of the water quality parameters like temperature, DO, pH
and turbidity from the year 2014 to 2019(till 18th Feb) for Georgia State with an hour of
intervals. To remove the noisy and inconsistent data, the data cleaning, and data scaling
(normalization) are carried out along with the graphical presentations and exploratory
analysis. To improve the efficiency of the model, a few re-enactments and parameter
determination are completed. The model shows that the RMSE for the LSTM is 0.03 and the
prediction results are proper. Our model is fit for multivariate parameters and the predictions
are done in an hourly manner. The web application is built to make it user-friendly.
Visualization of water parameters gives good insight for understanding the parameter
changes for past hours.
Future Scope
Developing a mobile app is more convenient for alerting the aquafarmers when critical
conditions exceed water parameter values and the app is accessible through phones. Giving
suggestions for aquafarmers based on seasons how to maintain water quality for the
surveillance of fishes. Water quality monitoring for aquaculture using IoT and predicting the
water quality using ML.

327
 Prediction of Water Quality System for Aquaculture using Machine Learning

References

Juntao Liu, Chuang Yu, Zhuhua Hu,” Automatic and Accurate Prediction of Key Water
Quality Parameters Based on SRU Deep Learning in Mariculture” 2018
IEEE International Conference on Advanced Manufacturing (IEEE ICAM 2018)
Wijayanti Nurul Khotimah,” Aquaculture Water quality prediction using Smooth SVM”,
IPTEK, Journal of Proceeding Series, Vol. 1, 2014 (eISSN: 2354-6026).
Sundarambal, P., Shie-Yui, L., & Pavel, T. (2009). Development of a neural network for
dissolved oxygen in seawater. Indian Journal of Marine Science, 38 (2), 151-159.
Lee, Y. J., & Mangasarian. (2001). A smooth support vector machine. Journal of
Computational Optimization and applications, 20: 5 – 22.
J. Wang, F. Xiang, F. Qiu, H. Wang, H. Liu. Research progress of “River Water Quality
Parameters Prediction Method Based on LSTM-RNN Model” Guild of Environmental
Sciences, Vol.37, No.4, 63-67, 2018.
H. B. Wang, Z. P. Yang, Q. Yu, T. J. Hong, X. Lin. Online reliability time series
prediction via convolutional neural network and long short-term memory for service-
oriented systems. Knowledge-Based Systems, Vol.159, 132-147, 2018.
Dong Yao, Lei Cheng, QiuXuan Wu, Gong Zhang, Bei Wu, YuQing He, “Assessment and
Prediction of Fishery water quality using electrochemical sensor array carried by
UAV”, Engineering Research Center of Metallurgical Automation and
Measurement Technology of Ministry of Education, Wuhan University of Science and
Technology, Wuhan, China
Abe, S. (2010). Support Vector Machines for Pattern Classification 2nd edition. New York:
Springer- Verlag.

328