© 2023 JETIR April 2023, Volume 10, Issue 4 www.jetir.

org (ISSN-2349-5162)

Water Quality Prediction Using Machine Learning

Algorithms

Dr. Poosapati Padmaja1, Ch. Siva Datta Sai2, V. Krishna Teja2, A. Prathyusha Ragav2,
P. Babji2
1 Professor, Department of Information Technology, Anil Neerukonda Institute of Technology and Sciences,
Sangivasala, Visakhapatnam, Andhra Pradesh, India.

2 Department of Information Technology, Anil Neerukonda Institute of Technology and Sciences,

Sangivasala, Visakhapatnam, Andhra Pradesh, India.

Abstract: One of the most essential components for the continuation of life is water, which accounts for about
70 percent of the surface area of the world. Water quality has deteriorated at an alarming pace due to rapid
urbanisation and industrialisation, which has led to the spread of devastating illnesses. The current concept of
real-time monitoring is useless since water quality has always been determined via costly and time-consuming
laboratory and statistical analysis. Poor water quality has major consequences; therefore, we need a solution that
is both effective and affordable. This work was motivated by the desire to better understand how to estimate the
water quality index (WQI), a single index for expressing water's overall quality, and the water quality class
(WQC), a distinct class generated from the WQI. The proposed method requires four input parameters: turbidity,
pH, trihalomethanes, organic carbon, hardness as CaCO3, sulphate, chloramines, conductance, and total
dissolved solids. The minimum absolute errors (MAEs) of 1.9642 and 2.7273 are achieved via gradient boosting
(with a learning rate of 0.1) and polynomial regression, respectively, when predicting the WQI. The highest
performance, however, is shown when categorising the WQC using the Random Forest (3, 7) configuration, with
an accuracy of 0.8507. The proposed method delivers excellent accuracy with a low number of parameters,
making it practical for use in real-time assessment of water quality systems.

Keywords: Water Quality Prediction; Supervised Machine Learning; Gradient Boosting; Random Forest.

JETIR2304287 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org c711
© 2023 JETIR April 2023, Volume 10, Issue 4 www.jetir.org (ISSN-2349-5162)
1. Introduction

Water is the most crucial resource since it is required to maintain all forms of life, yet it is constantly threatened
by pollution from existing organisms. One of the most pervasive and far-reaching communicative media is water.
As a direct result of rapid development, water quality has been rapidly declining. One of the main causes of the
spread of dangerous illnesses is poor water quality. Waterborne infections are responsible for 80% of all diseases
in underdeveloped nations, causing 6 million deaths and 2.6 billion illnesses per year [1]. Some of the most
common ones are diarrhoea, typhoid, gastritis, clostridium difficile infections, some forms of hepatitis, and
giardiasis intestinal worms [2]. Annually, 0.6-1.44% of India's GDP is lost due to waterborne infections [3]. This
makes it an urgent issue everywhere, but especially in a growing nation like India. The existing techniques of
evaluating water quality based on laboratory and statistical studies are ine ective [4] because water is such a
quickly transmissible medium and time is of the importance if water is contaminated with trash that causes
sickness. This requires a substantial investment of time and effort on collection, transportation, and computation
of samples.

The disastrous effects of water contamination need an expedient and affordable solution. The primary goal of
this research is to propose and assess a supervised machine learning-based alternative approach to real-time
water quality prediction. This study uses data from the Rawal watershed in India, which was collected by the
India Council for Scientific and Industrial Research to determine whether or not the water there is fit for human
consumption 1 for potable and 0 for not potable. (“https://www.kaggle.com/datasets/adityakadiwal/water-
potability /”). The water quality index (WQI) and water quality class were predicted using a collection of typical
supervised machine learning algorithms applied to the aforementioned dataset (WQC).

2. Literature Review

This study investigates the approaches used to date in an effort to address water quality issues. Some
investigations utilise machine learning approaches to help in identifying an optimal solution to the water quality
issue, while others rely on more traditional lab analysis and statistical analysis. We were able to better understand
India's water quality issue thanks to local studies that made use of laboratory analyses. One such study was
conducted by Daud et al. [5], who collected water samples from several locations throughout India and subjected
them to manual lab examination, finding an abundance of E. coli and faecal coliform owing to industrial and
sewage waste. In a manual lab examination, Alamgir et al. [6] discovered significant levels of sulphates and total
faecal coliform count in 46 samples from Orangi town, Karachi.

Following becoming acquainted with the water quality research in India, we investigated the water quality study
using machine learning techniques. Classical machine learning methods such as “Support Vector Machines
(SVM), Neural Networks (NN), Deep Neural Networks (Deep NN), and k Nearest Neighbors (kNN) have been
used to evaluate water quality, with the best accuracy achieved by Deep NN (93%)”.

The “World Health Organization (WHO) standards (Available online at URL

https://www.who.int/airpollution/guidelines/en/)” are used to evaluate the water quality in their job, and these
standards test for turbidity, temperature, and pH. Predictions of water quality are severely constrained when just
three characteristics are used and compared to standard values. To estimate the WQI, Ahmad et al. [8] used both
standalone feed-forward neural networks and multi-network ensembles. They fed in data on 25 indicators of
water quality. The MSE was reduced from 0.1158 to 0.1200, and the R2 was raised from 0.9270, all thanks to
the use of backward eradication and the forward choice selective combination approaches. Due to the use of 25
parameters and the price of the parameter sensors, their approach is a little excessive in terms of a cheap real-
time system. Sakizadeh [9] predicted the WQI using ANN with Bayesian regularisation and 16 parameters
relating to water quality.

The research found that the observed values were highly correlated with the anticipated values (0.94 and 0.77,
respectively). Using ANN and multivariate linear regression, two standard machine learning techniques,
JETIR2304287 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org c712
© 2023 JETIR April 2023, Volume 10, Issue 4 www.jetir.org (ISSN-2349-5162)
Abyaneh [10] made predictions about “the chemical oxygen demand (COD) and the biochemical oxygen demand
(BOD)”. For this purpose, they calculated the COD and BOD using four variables: pH, temperature, TSS, and
TS.

Samples were classified into water quality classes using the average linkage (within groups) approach of
hierarchical clustering, an unsupervised methodology developed by Qamar and Ali [11]. They did not apply a
standardised water quality indicator in their evaluations of the models' performance, and they did not take into
account the most important characteristics related to WQI in the course of their training. The WQI was predicted
by ANN in a study by Gazzaz et al. [4], with the resulting model accounting for over 99.5% of the observed
variance. They employed 23 characteristics to forecast the WQI, which given the costs of sensors, is too costly
for application in an IoT system.

Using a feedforward neural network, Rankovic et al. [12] made DO predictions (FNN). Again, defeating the
purpose if it is to be utilised for real-time WQI estimate with an IoT system is the usage of 10 factors to anticipate
the DO.

Most analyses either did not utilise an estimated water quality index standard, relied on manual lab analysis, or
included too many parameters. Figure 1 depicts the current technique that the proposed methodology improves
upon.

2019 11 x WATER PUBLIC REVIEW Predicting water quality with just three characteristics and comparing
them to standard values is fairly limited. To estimate the WQI, Ahmad et al. [8] used both standalone feed-
forward neural networks and multi-network ensembles. They fed in data on 25 indicators of water quality.

By using both backward elimination and forward selection selective combination techniques, they were able to
decrease the MSE from 0.1158 to 0.1200 and increase the R2 from 0.9270. Their method is a bit excessive in
terms of a cheap real-time system due to the utilisation of 25 parameters and the cost of the parameter sensors.
Sakizadeh [9] used ANN with Bayesian regularisation and 16 water quality factors to make predictions about
the WQI. A link was found via his research.

Most studies either did not estimate the water quality index standard, relied on manual lab analysis, or utilised
too many parameters. Figure 1 depicts the current technique that the proposed methodology improves upon.

Figure 1. Methodology flow.

JETIR2304287 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org c713
© 2023 JETIR April 2023, Volume 10, Issue 4 www.jetir.org (ISSN-2349-5162)
3. Data Preprocessing
Information on how the data used in this study was cleaned using a box plot analysis can be found in the next
section. Once the data had been cleansed, q-value normalisation was used to bring them within the range of 0-
100, where they could be used in the calculation of the WQI with its six input parameters. After arriving at the
WQI, we used z-score normalisation to make all raw data comparable. Below we will go through the whole
process in detail.

3.1. Data Collection

Six hundred sixty-three samples from 13 Rawal Water Lake sources were included in the PCRWR dataset
collected between 2009 and 2012. Table 1 details the 12 parameters it measured in conjunction with 51 samples
from each source.
3.2. Boxplot Analysis and Outlier Detection
Since most of the parameters varied considerably and clustered around the extremes of the range, and because
boxplot analysis provides helpful visualisation for setting outlier detection criteria, we used it. Based on the
results of the boxplot analysis, we determined that most parameters were outside of the box and hence outliers
were considered to be normal. Parameters with large deviations from the mean were replaced with the maximum
threshold value after being identified. The maximum threshold was set at the parameters that was just below the
outlier values. As can be seen in Figure 2, we set the turbidity threshold at 754, applying it to all readings over
that point and assigning them the number 7554 Due to the limited size of our data set, we ran the analysis many
times with different parameter settings and selected the outliers by hand [4]. Moreover, in order to avoid skewing
the dataset, we were very forgiving in setting the upper threshold of parameters, instead just lightly penalising
the values that were exceedingly implausible.

Figure 2. Box plot analysis is used to find outliers.

3.3. Water Qualiity Index (WQI)
The “Water Quality Index (WQI)” is a single number that may be calculated from a variety of factors that are
excellent predictors of water quality. While nine water quality parameters are normally used for estimating the
WQI, we may still do it with at least six of them. In our data set, we measured pH, turbidity, Hardness as CaCO3,
Organic Carbon, and total dissolved solids. We also took into account the weight and relative relevance of
chloramines, the sixth parameter in the WQI. Equation (1) shows how we used these parameters and their weights
to determine the WQI for each sample, where qvalue is the value of a parameter from 0 to 100 and w_ f actor is
the weight of a specific parameter taken from Table 2. First, we multiply the q value of each parameter by its
associated weight, then we add up all of those products, and finally we divide that total by the total weights of
the parameters we used [14,15].

JETIR2304287 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org c714
© 2023 JETIR April 2023, Volume 10, Issue 4 www.jetir.org (ISSN-2349-5162)
Table 2. Weights assigned to WQI parameters [14,15]

Weighing Factor Weight

pH 0.11

Turbidity 0.08

Total Dissolved Values 0.07

Chloromines 0.10

Organic Carbon 0.16

3.4. Water Qulaity Class (WQC)

Using the WQI as a descriptor in classification algorithms [14,15], we assigned each sample to a water quality
class (WQC) as indicated in Table 3.

Table 3. Ranges [14,15].

Water Quality Index Range Class

0-25 Very Bad

25-50 Bad

50-70 Medium

70-90 Good

90-100 Excellent

3.5. Q-Value Normalization

In order to facilitate index computation, the parameters, especially those pertaining to water quality, were
normalised using Q-value normalisation and placed in a range from 0 to 100. Six of the water quality indicators
are shown on q-value plots in Figure 3. Using these, we were able to scale five of these characteristics from their
original 0–100 range [14,15]. The q-value ranges for the sixth parameter, Chloromines, were unavailable,
therefore we utilised the WHO standards to clearly translate them to the 0-100 range using a set of thresholds as
follows: 100 for q1, 80 for q2, 50 for q3, and 0 for q4, representing severe punishment. After q-normalizing the
values such that they fell within the range of 0-100, they were used to determine the dataset's WQI using (1).

JETIR2304287 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org c715
© 2023 JETIR April 2023, Volume 10, Issue 4 www.jetir.org (ISSN-2349-5162)

Figure 3. Q-value normalization range charts.

3.6. Z-Score Normalization

One common approach to normalising and standardising data is the z-score to express the number of standard
deviations by which a given raw data point deviates from the population mean. Its optimal range is -3 to +3. To
make all the data on different scales fit within the standard one, it "normalises" the dataset to the one specified
above.

We used the z-score to normalise the data, which produces a score ideally spanning from -3 to +3, and is
calculated by Equation (2), where x indicates the value of a given sample, the mean, and the standard deviation,
respectively.

3.7. Data Analysis:

When the data was cleaned and processed, a number of machine learning techniques were applied to it in order to
make predictions about the WQI and WQC using as few factors as feasible. Before the machine learning algorithms
can be employed, the data that will be used as input must undergo some preliminary processing, such as correlation
analysis and data splitting.

3.7.1. Correlation Analysis : In order to find the dependant variables and make predictions for the difficult-to-
estimate variables using the easily-obtainable data, we performed correlation analysis to discover the probable
linkages between the parameters. The Pearson correlation, the standard and most trusted method for establishing a
link between two sets of data, was used in this study. The Pearson correlation was employed on the unnormalized
values of the parameters listed in Table 4 after q-value normalisation was applied, as will be explained later.

The results of the correlation analysis led us to conclude that temperature, turbidity, pH, and total dissolved solids are
the best variables to use when trying to predict WQI. As the first three factors are very inexpensive, we focused on
them initially; if required, TDS will be included for evaluation of its influence on accuracy.

JETIR2304287 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org c716
© 2023 JETIR April 2023, Volume 10, Issue 4 www.jetir.org (ISSN-2349-5162)

3.7.2. Data Splitting–Cross Validation

The last step before deploying a machine learning model is to divide the provided data into subsets for training
and testing the model, and to calculate accuracy measures to demonstrate the model's performance. In this
research, we look at the practise of cross-validation data splitting.

When using cross validation, the data is partitioned into k parts, with k minus one of those parts serving as the
training dataset and the remaining part serving as the testing dataset. This enables a fair data distribution during
training and testing, as well as a clean data split. In order to do cross validation, we partitioned the data into k =
6 subgroups. As there are a total of 663 samples in the training set, we made sure there were 100 in each fold
subset.

JETIR2304287 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org c717
© 2023 JETIR April 2023, Volume 10, Issue 4 www.jetir.org (ISSN-2349-5162)
3.7.3. Machine Learning Algorithms

Binary classification and logistic regression were both used in our research. Classification techniques were
utilised to sort data into the previously specified WQC, while regression procedures were employed to estimate
the WQI. We used 10 classification methods and 8 regression techniques. In our research, we used the following
algorithms:

1 Random forest is a model that makes judgements based on all of the models it has used to analyse subsets
of the provided data. The decision tree serves as the foundation for random forest, which combines the benefits
of utilising many models with those of using a decision tree[19].

2 SVMs, have a place in regression analysis in addition to their frequent usage in categorization. SVMs
provide a hyperplane between the classes and a larger margin for error so that data points projected onto a plane
may be more clearly shown and fewer false positives are produced [21].

3 The most cutting-edge algorithm employed in most contests now is the Gradient Boosting Algorithm. A
differentiable loss function may be optimised using an additive model. The 'ls' loss function, a min samples split
of 2, and a learning rate of 0.1 were the settings we found useful [20].

4 The K closest neighbour method determines a point's classification by determining which of its N nearest
neighbours belongs to which category. Many methods exist for breaking a tie, such as increasing the sample size
(n) or favouring one group over another. In each iteration, it takes the whole training set and finds the closest
neighbours [30]. In our simulation, n was set to 5.

5 For both classification and regression, you may rely on the straightforward decision tree technique.
Decisions about the various parameters are organised in a tree structure from the top down, and various outcomes
are predicted depending on the values of those parameters [31].

4. Results

Before diving into the findings, we'll go through the various metrics we utilised to evaluate the performance of
the machine learning techniques we put into play.

4.1. Accuracy Measures

As was previously indicated, this study made use of both regression and classification supervised machine
learning techniques. Both kinds of algorithms were given various grades based on the outcomes they produced.

The following variables were utilised for regression analysis:

i.Accuracy: A model's accuracy is measured by the proportion of its accurate predictions throughout the whole
data set. Equation (9) measures accuracy, where (TP, TN, FP, FN) denote true positive, true negative, false
positive, and false negative, respectively [7,35].

ii.Precision:
It measures how many occurrences of a positive class were properly categorised out of the total
number of instances of that class. Equation (10) is used to determine accuracy, where TP and FP are the
proportion of correct diagnoses and misdiagnoses, respectively [7,35,36].

JETIR2304287 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org c718
 © 2023 JETIR April 2023, Volume 10, Issue 4 www.jetir.org (ISSN-2349-5162)

iii.Recall:The recall measure indicates how many occurrences of a positive class were properly labelled. In the
equation (11), TP stands for true positive and FN for false negative [7,35,36]. This allows us to determine the
recall rate.

iv.F1score: We used the harmonic mean of precision and recall to represent the F1 score, as stated in Equation
(12), since these two metrics alone do not capture the whole scope of accuracy. Quantity is between zero and
one. Accuracy increases as the score rises [7,35,36].

JETIR2304287 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org c719
© 2023 JETIR April 2023, Volume 10, Issue 4 www.jetir.org (ISSN-2349-5162)
4.2. Results for Regression Algorithms

This research sought to forecast water quality using a minimum amount of factors and inexpensive sensors due
to the high cost of sensors measuring water quality parameters. Originally, we relied on four indicators: hardness
(in terms of CaCO3), turbidity, pH, and total dissolved solids. Table 5 displays the results of our application of
several “regression techniques; we found Gradient Boosting to be the most effective, with an MAE of 1.9642,
an MSE of 7.2011, an RMSE of 2.6835, and an RSE of 0.7485”.

Table 5. Regression results using four parameters.

Algorithm MAE MSE RMSE R Squared

Random Forest 2.3053 9.5669 3.0930 0.6705

Gradient Boosting 1.9642 7.2011 2.6835 0.7485

SVM 2.437 10.6333 3.2609 0.3458

After that, we sought to cut down on some additional metrics by eliminating total dissolved solids, which is
harder to collect than the others. Table 6 displays our findings showing “the SVM is the most effective method,
with a mean absolute error of 2.8253, a mean squared error of 13.8546, and a root mean squared error of 3.7222.
The total mistake rate was up, but it was still well below acceptable bounds and well worth the investment”.

Table 6. Regression results using three parameters.

Algorithm MAE MSE RMSE

Random Forest 3.0404 15.2473 3.9048

SVM 2.8252 13.8546 3.7222

Gradient Boosting 2.8060 13.2710 3.6429

4.3. Results for Classification Algorithms

Samples were categorised into water quality classes (WQCs) based on their WQIs, which were anticipated using
classification algorithms. Classification was performed using the same criteria as before. At first, the same four
factors were taken into account. Table 7 displays our findings that Random Forest outperformed the other
algorithms in this scenario, with a “reliability of 0.8506, precision of 0.5658, recall of 0.5641, and F1 score of
0.5648”.

Table 7. Results of a classification scheme based on four variables.

Algorithm Accuracy Precision Recall F1 Score

Gradient Boosting Classifier 0.8130 0.5375 0.5376 0.5376

KNN 0.7270 0.4734 0.4783 0.4750

Decision Tree 0.7949 0.5298 0.5250 0.5268

Random Forest 0.8506 0.5658 0.5641 0.5648

SVM 0.7979 0.5187 0.5327 0.5228

5. Conclusions and Future Work

WQI is used to assess the quality of water, which is a crucial resource for human life. Water quality testing often
entails a time-consuming and costly laboratory examination. This study investigated a machine learning
JETIR2304287 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org c720
© 2023 JETIR April 2023, Volume 10, Issue 4 www.jetir.org (ISSN-2349-5162)
technique that may be used to make predictions about water quality using just a few readily accessible factors.
“Data for the research came from the PCRWR and included 663 samples collected from 12 locations around
Rawal Lake in Pakistan. In order to estimate WQI, we used a collection of widely used supervised machine
learning methods. Results showed that when it came to predicting WQI, polynomial regression with a degree of
2 and gradient boosting with a learning rate of 0.1 performed better than other regression algorithms, while multi-
layer perceptron (MLP) with a configuration of (3, 7) performed better when it came to classifying WQC. We
recommend incorporating this study's results into a large-scale IoT-based online monitoring system in future
works, whereby only sensors for the necessary parameters would be used to collect data. Based on real-time data
received by the IoT system, the evaluated algorithms would quickly estimate the water quality”. The pH,
turbidity, and temperature sensors would be used in the proposed IoT system.

References
1. PCRWR. National Water Quality Monitoring Programme, Fifth Monitoring Report (2005–2006); Pakistan Council
of Research in Water Resources Islamabad: Islamabad, Pakistan, 2007. Available online:
http://www.pcrwr.gov.pk/Publications/Water%20Quality%20Reports/Water%20Quality% 20Monitoring%20Report%202005-06.pdf
(accessed on 23 August 2019).
2. Mehmood, S.; Ahmad, A.; Ahmed, A.; Khalid, N.; Javed, T. Drinking Water Quality in Capital City of Pakistan. Open Access
Sci. Rep. 2013, 2. [CrossRef]
3. PCRWR. Water Quality of Filtration Plants, Monitoring Report; PCRWR: Islamabad, Pakistan, 2010. Available online:
http://www.pcrwr.gov.pk/Publications/Water%20Quality%20Reports/FILTRTAION%20PLANTS% 20REPOT-CDA.pdf (accessed on
23 August 2019).
4. Gazzaz, N.M.; Yusoff, M.K.; Aris, A.Z.; Juahir, H.; Ramli, M.F. Artificial neural network modeling of the water quality index
for Kinta River (Malaysia) using water quality variables as predictors. Mar. Pollut. Bull. 2012, 64, 2409–2420. [CrossRef]
5. Daud, M.K.; Nafees, M.; Ali, S.; Rizwan, M.; Bajwa, R.A.; Shakoor, M.B.; Arshad, M.U.; Chatha, S.A.S.; Deeba, F.; Murad,
W.; et al. Drinking water quality status and contamination in Pakistan. BioMed Res. Int. 2017, 2017, 7908183. [CrossRef]
6. Alamgir, A.; Khan, M.N.A.; Hany; Shaukat, S.S.; Mehmood, K.; Ahmed, A.; Ali, S.J.; Ahmed, S. Public health quality of drinking
water supply in Orangi town, Karachi, Pakistan. Bull. Environ. Pharmacol. Life Sci. 2015, 4, 88–94.
7. Shafi, U.; Mumtaz, R.; Anwar, H.; Qamar, A.M.; Khurshid, H. Surface Water Pollution Detection using Internet of Things. In
Proceedings of the 2018 15th International Conference on Smart Cities: Improving Quality of Life Using ICT & IoT (HONET-ICT),
Islamabad, Pakistan, 8–10 October 2018; pp. 92–96.
8. Ahmad, Z.; Rahim, N.; Bahadori, A.; Zhang, J. Improving water quality index prediction in Perak River basin Malaysia through
a combination of multiple neural networks. Int. J. River Basin Manag. 2017, 15, 79–87. [CrossRef]
9. Sakizadeh, M. Artificial intelligence for the prediction of water quality index in groundwater systems.
10. Model. Earth Syst. Environ. 2016, 2, 8. [CrossRef]

11. Abyaneh, H.Z. Evaluation of multivariate linear regression and artificial neural networks in prediction of water quality
parameters. J. Environ. Health Sci. Eng. 2014, 12, 40. [CrossRef]
12. Ali, M.; Qamar, A.M. Data analysis, quality indexing and prediction of water quality for the management of rawal watershed in
Pakistan. In Proceedings of the Eighth International Conference on Digital Information Management (ICDIM 2013), Islamabad,
Pakistan, 10–12 September 2013; pp. 108–113.
13. Rankovic´, V.; Radulovic´, J.; Radojevic´, I.; Ostojic´, A.; Cˇ omic´, L. Neural network modeling of dissolved oxygen
14. in the Gruža reservoir, Serbia. Ecol. Model. 2010, 221, 1239–1244. [CrossRef]
15. Kangabam, R.D.; Bhoominathan, S.D.; Kanagaraj, S.; Govindaraju, M. Development of a water quality index (WQI) for the Loktak
Lake in India. Appl. Water Sci. 2017, 7, 2907–2918. [CrossRef]
16. Thukral, A.; Bhardwaj, R.; Kaur, R. Water quality indices. Sat 2005, 1, 99.
17. Srivastava, G.; Kumar, P. Water quality index with missing parameters. Int. J. Res. Eng. Technol. 2013, 2, 609–614.

JETIR2304287 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org c721