Early Detection of Fish Diseases by Analyzing Water Quality Using Machine Learning Algorithm
Al-Akhir Nayan, Ahamad Nokib Mozumder, Joyeta Saha, Khan Raqib Mahmud, Abul Kalam Al Azad
IInternational Journal of Advanced Science and Technology Vol. 29, No. 5, (2020), pp. 14346 - 14358
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
Early Detection of Fish Diseases by Analyzing Water Quality Using Machine Learning Algorithm
Al-Akhir Nayan , Ahamad Nokib Mozumder , Joyeta Saha , Khan Raqib Mahmud , Abul Kalam Al Azad Lecturer, Department of Computer Science & Engineering, European University of Bangladesh, Dhaka, Bangladesh Lecturer, Department of Computer Science & Engineering, University of Liberal Arts Bangladesh, Dhaka, Bangladesh Associate Professor, Department of Computer Science & Engineering, University of Liberal Arts Bangladesh, Dhaka, Bangladesh (*Corresponding author’s e-mail: [email protected])
Abstract
Early detection of fish diseases and identifying the underlying causes are crucial for farmers to take necessary steps to mitigate the potential outbreak, and thus to avert financial losses with apparent negative implications to national economy. Typically, fish diseases are caused by virus and bacteria; according to biochemical studies, the presence of certain bacteria and virus may affect the level of pH, DO, BOD, COD, TSS, TDS, EC, PO43-, NO3-N, and NH3-N in water, resulting in the death of fishes. Besides, natural processes, e.g., photosynthesis, respiration, and decomposition also contribute to the alteration of water quality that adversely affects fish health. Being motivated by the recent successes of machine learning techniques in complex relational data analyses in accurate classification and decision-making tasks, a state-of-art machine learning algorithm has been adopted in this paper to detect and predict the degradation of water quality timely and accurately, thus it helps taking pre-emptive steps against potential fish diseases. The experimental results show a high accuracy in detecting fish diseases particular to specific water quality based on the algorithm with real datasets.
Keywords: water quality analysis, water quality prediction, diseases identification, bacteria attack, automatic detection, gradient boosting. Introduction
Fishes account for approximately 15% of the animal protein intake of the human population globally [1]. In countries like Bangladesh, fishes provide as high as 60% of the animal protein to the populace, also in economic valuation fishes contribute to approximately 3.6% to the national GDP which weighs nearly 25% of the entire agricultural GDP [2]. Furthermore, this section employs about 11% of the total population in Bangladesh in full-time and part-time basis [2]. Despite being the very vibrant economic sector for the country, one major threat to the fish farmers are the fish diseases which eventually puts a huge constraint on the economic progress, and severely strains the expansion of the aquaculture and fish farming [3]. Fish culture faces severe threat from waterborne pathogens, such as bacteria and virus, responsible mainly for mass mortality and poor health. It is, therefore, imperative to monitor the purity of the water habitat to detect fish diseases timely and accurately. nternational Journal of Advanced Science and Technology Vol. 29, No. 5, (2020), pp. 14346 - 14358
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
Fish performs all their physical activities under water; fish dependents on water for breathing, feeding, reproducing and growth. When the water quality of the habitat deteriorates it becomes unfavorable for fish to live in. Water quality depends on certain parameters, and when the parameters change the quality deteriorates. As a result, the health of fishes is threatened by the compromise of their immune system critically leading them to be vulnerable to harmful pathogens [4]. Oxygen plays a pivotal role in maintaining life under water. When oxygen level goes below to the preferable range, the physiological and physical growth of fish species are hampered. Decreasing oxygen level under water results in increasing carbon dioxide level causing acidosis and nephron calcinosis that obstructs the development of granulomas in many internal organs of fish [5, 6]. Moreover, recirculated water contains high pH which turns up ammonia level in water. High level of ammonia causes harm to the gills and liver of fishes. Depending on the level of saturation and the time of exposure, Gas supersaturation of the water can result in the gas bubble disease. The main cause to the disease is the development of bubbles in the eyes, skin and gills [7]. Degraded water quality causes pollution that creates serious problem to fish; necrotic alteration, papilloma, degenerative, and fin erosion is the result of water pollution. As a result, the fish body gets abnormal growth, and farmers do not get optimum production. Problems such as these can easily be resolved if the farmers could identify them early. Machine learning techniques and artificial intelligent algorithms in recent years have been used extensively and very successfully in classification and decision-making problems [8, 9]. Smart algorithms can learn from parameter space of system the correct desired classification based on real dataset, and infer very accurately any deviation from the desired sets of configuration, which may be exploited for decision-making. A hierarchical architecture to the algorithm ensures higher performance from learning. Machine learning technique, such as Gradient boosting, exploits decision-tree based hierarchical in-built structure based on regression algorithm to classify complex problems and to help in the decision-making automation [10, 11]. Inspired by the successes of such technique in variety of complex problems, we have employed the technique in this study to predict fish diseases by solving classification problems from real dataset composed of desired parameters of water purity. Our approaches to the development of an automated fish disease detection algorithm and employing it in decision-making process may be summarized in the following sequential steps: Step 1: Taking sample of water to identify the water quality. Step 2: Making prediction of the water quality using machine learning algorithm. We have already collected and prepared dataset and trained our algorithm using it, so that machine can make prediction on probable fish diseases based on the water quality parameters. Step 3: Analyzing the disease and identifying. Step 4: Making smart decision to minimize harm to fish farm and ensuring healthy habitat. The paper is arranged as following: the review of the related literature is conducted in the second section followed by an in-depth look into the proposed machine learning algorithm and parameters of the model in the section three. The dataset is discussed in the section four along with the preparation, processing, and implementation in the model. The experimental results are discussed in the section five followed by concluding remarks. nternational Journal of Advanced Science and Technology Vol. 29, No. 5, (2020), pp. 14346 - 14358
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC Related Works
In 2012, Chemical Oxygen Demand (COD), Total Suspended Solids (TSS) and Oil & Grease (O&G) were monitored using multi sensor water quality monitoring system. An improved boosting method was used for suppressing quality-relevant variables by applying smaller weights and constructing models based on weighted variables for wastewater quality predictions. The observing framework was tried in the field with good outcomes, hidden the capability of this method for the online checking of water quality [12]. Another study took place in 2014 to measure the pollution of South Korea Ocean using machine learning technique. They collected 63 samples from 2011 to 2012. The study successfully discussed the distribution of water quality parameters analyzing the Geostationary Ocean Color Imager (GOCI) images [13]. On 2012, a study by S. Shah was carried out on twenty different locations of Kerala, India. The study showed that the majority of water samples in the region were suitable for agriculture purposes and a very easy pretreatment was also necessary to make the water suitable for the lives under water [14]. N. Karlar collected water samples from ten villages and did analysis on it. He nalyzed physico-chemical variables like TH, Temp, F—, PH, Cl—,TDS, Ca2 +, Mg2 +, alkalinity and SO4 2-. The WQI of these specimens, ranged from 40.67 to 69.59, showed that koilwar block surface water was needed to be processed before use [15]. In 2013, Usha carried out research on the determination of the water quality rating and health of urban water bodies in Bilari town. Water samples were obtained from ten separate locations in the three months January, June and September 2011. Examination of surface water were done for various physico-synthetic parameters. WQI indicated that water contamination were rising day by day [16]. A. B. Frontier worked on his research. To analyze Physico-chemical properties of water, samples obtained from selected villages in Nasik district of Kalwan Tahsil. Water tests were gathered haphazardly from five distinct locations. Result shows that physico-compound parameters of gathered water tests are inside passable cutoff in study zone. So it could be said that water condition was favorable for the lives under water [17]. N. Bagde researched the assessment of water quality and its effects with specific reference to Chhindwara district. Water samples from 5 blocks of the infected area were obtained and examined for physico-chemical variables such as alkalinity, pH, total hardness, electrical conductivity and fluoride ion. Many water samples showed greater concentration of fluoride ions and greater turbidity. Most of the samples needed chemical treatment to make it suitable for the living body. Adeyemi et al. did their research on the chemical impact of leachate on the consistency of Nigeria's surrounding water. The analysis showed that in dry season the BOD and COD of the leachate-contaminated water samples were greater than in rainy season. Bacteria were more in rainy season than dry season. Consequently, the data of this study showed that use of leachate-contaminated water was harmful and should be prevented [18]. Physical chemical experiments like pH, dissolved oxygen chloride, temperature, total alkalinity, , total dissolved solids, calcium & sulphate, magnesium hardness , phosphate and bore well water nitrate were conducted in 20 villages of the town of Gondpipri and some of its interior adivasi area during 2010, in order to evaluate the water quality index. The findings show that pH & TDS is within the same limits that WHO & ISI provides. Calcium, Mg, Sulphate are within permissible limits. But the levels of phosphate, nitrates are higher values which prescribe values which is harmful for fish farming [19]. nternational Journal of Advanced Science and Technology Vol. 29, No. 5, (2020), pp. 14346 - 14358
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
In another study, Physico-chemical parameters such as temperature, pH, sulphate, electrical conductivity, bicarbonate, sodium, calcium, potassium, magnesium, iron, silica ,chloride, nitrate, dissolved oxygen, phosphate, biological demand for oxygen & chemical demand for oxygen were monitored. From 2017 February to 2018 March samples of those parameters were collected and analyzed from 8 separate reservoir locations. From this analysis, it is shown that there is a frequent variety of convergence of physico-chemical parameters and some of the parameters are past passable breakpoints, showing water quality degradation due to contamination [20] Most of the studies indicate the change of water quality due to the change of the chemical components of water. Inspiration drawn from these studies has led to the current investigation on water quality governed by the parameters based the chemical composition of water. And the study further links this degradation of water to the emergence of various fish diseases. Research Method
Gradient boost is a sophisticated machine learning tool used for solving regression and classification problems in complex decision-making tasks. The model is comprised of computational stages made up of decision trees. The model executes a given classification task based on minimizing a differentiable loss function in every computational stages of the architecture. The stages are coupled with in the learning rule with a weak learning rate in an additive formalism. The underlying feature of this functional architecture is that as the model learns from every successive computation, the performance gets better gradually as it learns to minimize its mistake by gauging the loss function. Thus, the model manages to learn the complex inter-relationship among the parameters and variables in small steps, as if the algorithm matures gradually with every model generation. The working procedure of the model has been shown in the
Figure 1 . Algorithm 1: Gradient Boost Algorithm
Inputs: Data input (𝑥, 𝑦) 𝑖=1𝑁 Iterations number 𝑀 The loss-function
Ψ (𝑦, 𝑓) The base-learner model (𝑥, 𝜃)
Algorithm: Initialize 𝑓 with constant 2. For 𝑡 = 1 to 𝑀 do Negative gradient computation 𝑔 𝑡 (𝑥) Fit a new base learner function (𝑥, 𝜃 𝑡 ) Find the best gradient decent step-size 𝜌 𝑡 : nternational Journal of Advanced Science and Technology Vol. 29, No. 5, (2020), pp. 14346 - 14358 ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC 𝜌 𝑡 = arg 𝑚𝑖𝑛 𝜌 Ψ 𝑁𝑖=1 [𝑦 𝑖 , 𝑓 𝑡−1 𝑥 𝑖 + 𝜌(𝑥 𝑖 , 𝜃 𝑡 )] Update the function estimate: 𝑓 𝑡 ← 𝑓 𝑡−1 + 𝜌 𝑡 (𝑥 𝑖 , 𝜃 𝑡 ) END
Figure 1. Working Procedure of Gradient Boosting
The numeric values of the model parameters used in the simulation are shown in the
Table 1.
Table 1. Numeric Value of Model Parameters
Name of Parameter
Value
Minimum Sample Split
Learning Rate
Max Depth Minimum Sample Leaf Model Parameters and Description
The
Table 2 below shows the computational types of the parameters used in the model for calculation and prediction of water quality.
Table 2. Model Parameters
Parameter Name
Type
Unit
STATION CODE object
LOCATIONS object
STATE object
Temp float64
D. O. float64 mg/l pH float64 CONDUCTIVITY object mhos/cm
B.O. D. float64 mg/l
NITRATENAN N+ NITRITENANN float64 mg/l
FECAL COLIFORM object MPN/100ml
TOTAL COLIFORM float64 (MPN/100ml) Mean nternational Journal of Advanced Science and Technology Vol. 29, No. 5, (2020), pp. 14346 - 14358
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
Year
Int64
Float64 has been taken because of storing fraction values for the parameters. In the table, the parameter Temp indicates the temperature of water. Station code, location and state provided the information about the place from where the sample data has been collected. D.O., which stands for Dissolved Oxygen, indicates the level of free and non-compound oxygen contained in water. This is an important parameter in the assessment of water quality due to its effect on the species within a water habitat. A too high or too low level of D. O. can damage aquatic life and affect water quality [21]. The pH indicates the amount of acid/basic compound in the water. Water pH regulates the solubility and biological quality of contaminants such as nutrients and heavy metals. In the case of heavy metals, toxicity is determined by the degree to which they are soluble. Conductivity measures the capability of passing electric current through water [22, 23] Biochemical Oxygen Demand (BOD) is a measure of the amount of dissolved oxygen used by aerobic microorganisms when the organic matter is decomposed in water. It offers an index for evaluating the impact of wastewater discharged on the recipient area. High BOD value indicates high amount of available organic compound for bacteria that consumes oxygen [24]. Nitrate is formed by combining oxygen or ozone with nitrogen. Nitrogen is useful for all living body. But higher level of nitrate is harmful for all the living body under water. The term Coliform indicates bacteria that are present in the animals including human. Coliform do not cause diseases but some coliform like E. coli can cause serious harm to living body [25, 26].
Water quality is determined by a variable WQI which is constructed based on the information of npH, NBDO, NEC, NNA, WPH, WDO, WBDO, WEC, WNA, and WCO. The variable is used to perform mathematical calcualtion as showing in Table 3.
Table 3. Calculation of Chemical Components in Water
Calculation Name
Component Range
Calculated Value
Calculation Name
Component Range
Calculated Value
Calculation of npH
PH Range npH Value Calculation of ndo DO Range ndo Value
Calculation of nco
TC Range Nco Value Calculation of nbdo BOD Range Nbdo Value
Calculation of nec
CO Range Nec Value Calculation of nna NA Range Nna Value
Now, nternational Journal of Advanced Science and Technology Vol. 29, No. 5, (2020), pp. 14346 - 14358
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC 𝑤𝑝 = 𝑛𝑝𝐻 ∗ 0.165 𝑤𝑑𝑜 = 𝑛𝑑𝑜 ∗ 0.281 𝑤𝑏𝑑𝑜 = 𝑛𝑏𝑑𝑜 ∗ 0.234 𝑤𝑒𝑐 = 𝑛𝑒𝑐 ∗ 0.009 𝑤𝑛𝑎 = 𝑛𝑛𝑎 ∗ 0.028 𝑤𝑐𝑜 = 𝑛𝑐𝑜 ∗ 0.281 𝑤𝑞𝑖 = 𝑤𝑝 + 𝑤𝑑𝑜 + 𝑤𝑏𝑑𝑜 + 𝑤𝑒𝑐 + 𝑤𝑛𝑎 + 𝑤𝑐𝑜
The value of the variable WQI measured as given is used in the learning algorithm to predict the water quality based on the constituting parameters; a certain value of the variable relates to a certain fish disease [27, 28].
A dataset was prepared for training the model. Approximately 2000 samples were collected from different places of Bangladesh. Before training the data was preprocessed. The
Table 4 provides information about the collected dataset. At the time of taking sample we collected other necessary information for our calculation process. All the terms used in the
Table 4 have already been described in the Method. The model can identify diseases and provide smart suggestion to prevent the diseases. Depending on the water quality range we have found out the probable disease and suggestions according to medical science. The
Table 5 provides a short description about the disease and the suggestions. We have trained our model with 63 diseases, their symptoms and the suggestions. 3D visual representation of filtered data has been shown to
Figure 2, while the
Figure 3 depicts the scatter plot of data, which is a 2D representation of
Figure 2 . The X axis of the
Figure 2 represents the time period in year and the Y axis represents the measure of water quality. We note from
Figures 2 and 3 that the water quality was better in 2017 than that of any other years.
Table 4. Dataset Configuration
Serial No
STATION CODE
LOCATIONS
State
Temp
D.O. (mg/l) pH CONDUCTIVITY
B.O.D.
NITRATENAN N+ NITRITENANN (mg/l)
FECAL COLIFORM (MPN/100ml)
Total COLIFORM (MPN/100ml) Mean
Month and year Table 5. Diseases with Relevant Reason and Suggestion
Serial
Disease
Reason
Suggestion Acid Death Low PH level Use chemical to increase Basic Compound Fungus in mouth Caused from bacterium Use of Penicillin nternational Journal of Advanced Science and Technology Vol. 29, No. 5, (2020), pp. 14346 - 14358
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC Growth Process Slow Lack of protein Protein Synthesis Tuberculosis Caused by the Bacterium Mycobacterium piscium Destroy infected fish Tail Rot & Fin Rot Caused by the bacteria Aeromonas Use CuSO4 … … … … Alkaline Death Decrease of hydroxyl ion in water Use chemical to increase acid compound in water Ulcer caused by baceria, haemophilus CUSO4 for one minute for a period of 3 to 4 days Ichthyosporidium Caused by fungus Add Phenoxethol to food Velvet or Rust Infection in gill a velvet by virus Copper at 0.2 mg per liter Nematoda Nematodes (threadworms) infect soak the food in para-chloro-meta-xylenol No Production Bacteria Attack Minimize acidity by using soda lime
Figure 2. 3D Visualization of Filtered Data
Figure 3. Planar Scatter Plot of Data Points nternational Journal of Advanced Science and Technology Vol. 29, No. 5, (2020), pp. 14346 - 14358
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
Standard pH meter, temperature, D.O, Conductivity, B.O.D. and Nitrate measurement instruments were used at the time of collecting data. For processing the data, we used Microsoft Excel. A core i5 laptop PC was used. The PC was configured with 2GB NVIDIA graphics memory with 2GB Intel graphics memory, 12 GB RAM and 1TB SSD.
The GBM (Gradient Boosting Model) library was installed and called to train the Gradient Boosting Model in R [29-31]. Specific arguments and calculations are needed for functioning the GBM. The necessary arguments and calculations are mentioned in method. Using the arguments, the model generates predictor variables. Then the response variables are specified. If it is not specified the model guesses a value. Finally, the data is specified. As GBM is an ensemble of decision-trees, the n-trees arguments are also specified. Then the training process gets started, and after training and testing the model provides results, and also the predictions. The training cost has been mentioned in the
Figure 4 . The loss function is minimized with the increasing number of iterations. At the starting time the loss was at the peak with high value, but when the training started it started to decrease until reaching a flat-line indicating the minimum value. The X axis of the graph indicates the number of iterations and Y axis represents the cost.
Figure 4. Cost Function Result and Discussion
The poor learner is built up in each training round, and predictions are measured with the right outcome. The difference among prediction and reality represents the model's error rate. The errors were used to calculate the gradient which was the loss function’s partial derivative. Direction finding was performed by the gradient. Following the training and testing, the model provided detailed analysis shown in Table 6. The model yields prediction depending on the analyzed result. The model can predict the water quality and also the diseases that are caused by the change of specific water quality features. A plot of actual result and predicted values are illustrated in
Figure 5 . nternational Journal of Advanced Science and Technology Vol. 29, No. 5, (2020), pp. 14346 - 14358 ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
The
Figure 5 has established a relationship among two factors, the data collection year and the WQI. Here the year, shown in the X axis, is working as independent variable and the WQI, shown in the Y axis, is acting as depended variable. The relationship between the two axis is approximated by the straight line. The line slants upward with the lower stopping point at the Y-catch of the diagram and the upper stopping point broadening upward into the diagram field, away from the X-axis. The two variables exhibit a positive linear relationship. The model fits with data very well. It has score Mean Square Error 1.1987755149740886 and R squared is 74.63% approximately 75%. The standard deviation of the blunders is actually one-portion of the standard deviation of the dependent variable.
Table 6. Data Analysis
S.N
Station
Location do ph co bod na tc year … nbdo nec nna wph wdo wbdo wec wna wco wqi … … … … … … … … … … … … … … … … … … … … … Figure 5. Plotting the Actual and Predicted Results
The
Table 7 is showing the predicted water quality, prediction errors, the effect on fish due to the change of WQI and the smart decision prepared by the model. The column nternational Journal of Advanced Science and Technology Vol. 29, No. 5, (2020), pp. 14346 - 14358
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC ―WQI‖ shows the present quality of a data. The ―Predicted WQI‖ column shows the prediction of the previous ―WQI‖ columns data. We collected samples from 500 different sources. From each source we collected 3 samples by 4 months interval and trained it to the model. The model predicts 4 months future by analyzing the given data. The ―Percentile Error‖ column judges the accuracy of the prediction. The model is trained with the possible diseases caused by the change of the WQI score and it takes smart decision against it. The column ―Chances of Disease‖ and ―Decision‖ are showing the machine generated output for the current change of WQI. The decision can be effective against the fish disease. The model can predict WQI and measure disease with an average of 92% accuracy.
Table 7. Prediction, Diseases Identification and Smart Decision
Serial
WQI
Predicted WQI (After 4 months)
Percentile Error
Chances of diseases
Decision Conclusion
This work presents a machine leaning based automated solution for early detection of fish diseases based on chemical contents analysis in water habitat. The proposed model incorporates a sophisticated classifier algorithm which is allowed to learn a real dataset to assess water quality and eventually to infer smart decisions. One major aspect of the research is the collection and processing of the dataset which is used for benchmarking the proposed model performance. The model is capable of producing realistic decision on actions against potential diseases in the habitat. The accuracy of the model is found to be very satisfactory. The early warning for the fish diseases will certainly be of great relief to the fish farmers as this model not only help to detect disease but also capable of referring to effective pre-emptive action to avert the disease. This study positions itself on a great potential to explore further extension to incorporate more types of diseases and mitigating decisions, and also IoT based implementation should be explored based on the proposed algorithm to equip fish farmers with necessary technology for quick and reliable on-site usage.
References [1]
DoF. 2018. ―Yearbook of Fisheries Statistics of Bangladesh, 2017-18. Fisheries Resources Survey System (FRSS).‖ Department of Fisheries. Bangladesh: Ministry of Fisheries, vol. 35, (2018), pp. 129. nternational Journal of Advanced Science and Technology Vol. 29, No. 5, (2020), pp. 14346 - 14358
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC [3]
N.Bagum, M. Monir and M.S. Khan., ―Present Status of Fish Diseases and Economic Losses Due to Incidence of Disease in Rural Freshwater Aquaculture of Bangladesh.‖ J. Innov. Dev. Strategy,vol. 7, no. 3, ( , 48-53. [4]
M.Chaurasia, I.Gupta, R.Kumar, P.Prasad and R. A.Sohony,―Water quality analysis of surface water: A Web approach.‖ Environmental Monitoring and Assessment, vol. 185, (2013) , pp. 5987–5992. [5]
N.Rahmanian, S. B.Hajar, M.Homayoonfard, N. J.Ali, M.Rehan, Y.Sadefand A. S.Nizami, ―Analysis of Physiochemical Parameters to Evaluate the Drinking Water Quality in the State of Perak, Malaysia.‖ Journal of Chemistry, (2015). [6]
C.Nagamani, C. S.Devi, and A.Shalini,―Physico-chemical analysis of water samples.‖ International Journal of Scientific & Engineering Research, vol. 6, no. 1, ( , pp. 2149-2155. [7]
M.Tanekhy,―Fish Health and Diseases.‖ ICMALPS 2015, Alexandria University, Egypt, (2015) , pp. 222-232. [8]
N.Emanet, H. R.Öz, N.Bayram and D. A.Delen,―Comparative analysis of machine learning methods for classification type decision problems in healthcare.‖ Decision Analytics, (2014) . [9]
P.Jeatrakuland K. W.Wong,―Comparing the performance of different neural networks for binary classification problems.‖ Eighth International Symposium on Natural Language Processing, ( ). [10]
K. L.Mosier, L. J.Skitka, S.Heers, and M.Burdick,―Automation Bias: Decision Making and Performance in High-Tech Cockpits.‖ The International Journal of Aviation Psychology,vol. 8, no. 1, (2009) , pp. 47-63. [11]
J. G.Harris and T. H.Davenport,―Automated decision making comes of age.‖ MIT Sloan Management Review, (2005) . [12]
C. M.Han, S. Y.Yang, H.Ceongand J. S.Park,―Clinical Sign-Based Fish Disease Diagnosis Aid System.‖ Electrical Engineering and Control, vol. 98, (2011), pp. 1039-1046 [13]
P.Mădălina And I. G.Breabăn,―Water Quality Index – An Instrument for Water Resources Management.‖ AERUL ŞI APA COMPONENTE ALE MEDIULUI, (2014), pp. 391-398. [14]
I. A.Ani,―Mathematical Computation of Water Quality Index for the Assessment of Al-Hilla River Ecosystem.‖ International Journal of Civil Engineering and Technology, vol. 10, no. 1, (2019), pp. 1862-1869. [15]
S.Chandra, S. S.Asadiand M. V. S.Raju,―Estimation of Water Quality Index by Weighted Arithmetic Water Quality Index Method: A Model Study.‖ International Journal of Civil Engineering and Technology, vol. 8, no. 4, (2017 ), pp.1215-1222. [16]
N. H.Omer,―Water Quality Parameters. Water Quality – Science.‖ Assessments and Policy, (2019) . [17]
A.Appavu, S.Thangavelu, S.Muthukannan, J. S.JesudossandB.Pand,―Study of Water Quality Parameters of Cauvery River Water in Erode Region.‖ Journal of Global Biosciences, vol. 5, no. 9, (2016) ,pp. 4556-4567. [18]
S. S.Sagar, R. P.Chavan, C. L.Patil, D. N.Shinde and S. S.Kekane,―Physico-chemical parameters for testing of water- A review.‖ International Journal of Environmental Sciences, vol. 3, no.3, (2012) , pp. 1194-1207. [19]
Z.Wu, D.Zhang, Y.Cai, X.Wang, L.Zhang and Y.Chen,―Water quality assessment based on the water quality index method in Lake Poyang: The largest freshwater lake in China.‖ Scientific Reports, vol.7, no. 1, (2017) . [20]
A.Natekinand A.Knoll,―Gradient Boosting Machines, A Tutorial.‖ Frontiers in Neurorobotics, (2013) . [21]
J. H.Friedman,―Stochastic Gradient Boosting.‖ Computational Statistics & Data Analysis, vol. 38, no. 4, (2002) , pp. 367-378. [22]
J. H.Friedman,―Greedy Function Approximation: A Gradient Boosting Machine.‖ Annals of Statistics,vol. 29, no. 5, (2001) , pp.1189-1232. [23]
X.Qin, F.Gao And G. Chen,―Wastewater quality monitoring system using sensor fusion and machine learning techniques.‖ Water Research, vol. 46, no. 4, (2012) , pp. 1133-1144. [24]
Y. H.Kim, J.Im, H. K.Ha., J. K.Choi and S.Ha,―Machine learning approaches to coastal water quality monitoring using GOCI satellite data.‖GIScience& Remote Sensing, vol. 2, (2014) , pp. 158-174. [25]
D. G. Shah and P. S. Patel,―Water quality of some drinking water samples of Kathalal territory.‖ Guidelines for Drinking-water Quality, 3rd ed. (2012) . [26]
S.Acharya, S. K.Sharmaand V.Khandegar,―Assessment of groundwater quality by water quality indices for irrigation and drinking in South West Delhi, India.‖ Data in Brief,vol. 18, ( ), pp. 2019-2028. nternational Journal of Advanced Science and Technology Vol. 29, No. 5, (2020), pp. 14346 - 14358
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC [27]
T. A.Idowu, H. A.Adedeji and O. A.Sogbesan,―Fish Disease and Health Management in Aquaculture Production.‖ Fresh Water Pollution Dynamics and Remediation, (2017) . [28]
U.Usha, S.Ahmad and K.Pramendra,―Determine water quality index and fitness of urban water bodies in Bilari town.‖ Journal of Chemical and Pharmaceutical Research, vol. 5, no. 11, (2013) , pp. 726-731. [29]
K.Neerja, K.Rajesh, S. S.Yadav and R. T.Singh,―Physico-chemical analysis of ground water taken five blocks (Udwantnagar, Tarari, Charpokhar, Piro, Sahar) of Southern Bhojpur (Bihar).‖ J. Chem. Pharm. Res, vol. 4,no. 3, (2012) , pp. 1827-1832. [30]
K.Rajesh and S. S.Yadav.―Correlation analysis of ground water quality in and around shahzad Nagar block of Rampur district, Uttar Pradesh, India.‖ Int. J. Chem. Sci, vol. 9,no. 1, (2011), pp. 440-447. [31]
S. S.Yadav and R.Kumar,―Assesment of physico–chemical properties of ground water taken from four blocks (Suar, Milak, Bilaspur, Shahabad) of Rampur district, Uttar Pradesh, India.‖RasayanJ. Chem, (2010),(2010),