The aim of our research is to calculate the Water Quality Index of bore water in our surrounding educational institutions using three learning algorithms. Our research work differentiates from other work by choosing Decision Tree, K-Nearest Neighbor, and Naive Bayes and analyzing their performance with accuracy. We collected water samples from various resources and calculated the six important factors: salinity, total suspended solids (TDS), dissolved oxygen (DO), acidity and alkalinity (pH), and biochemical oxygen demand (BOD). Using efficient chemical methods, the quality parameters of water were examined. We created our dataset by utilizing these metrics, and the dataset is given as our chosen algorithm’s training and testing data. We implemented these machine learning algorithms using Google Colab. Finally, we got the WQI value with three different accuracies.
Article
Analyzing the Performance of Machine Learning Algorithms for Predicting Water Quality Index
https://doi.org/10.21203/rs.3.rs-3879167/v1
This work is licensed under a CC BY 4.0 License
Version 1
posted
You are reading this latest preprint version
Decision Tree
Gini Index
KNN
Naive Bayes Water quality Index
One of the major resources for human beings is water. People use water frequently in their day-to-day lives. We should use pure water to avoid skin and lung diseases. For this purpose, we have calculated the value of the Water Quality Index [1] of water.
The methods of quality assessment of water differ in their methodologies as well as their input parameters.[1]. The most frequent Water Quality Index Methods are the National Sanitation Foundation Method, the Oregon Water Quality Index Method, the Weighted Arithmetic Water Quality Index Method, and the Canadian Council of Ministers of the Environment Water Quality Index Method[2]. We adopted the Weighted Arithmetic Water Quality Index Method in this research paper. We calculated the important parameters such as dissolved oxygen, salinity, acidity & alkalinity, total suspended solids (TDS), and biochemical oxygen demand (BOD) oxygen and tabulated it as a CSV file.
Nowadays many problems are efficiently solved by machine learning algorithms. The most important algorithms are Decision Tree, Support Vector Machine, Regression, Random Forest, and clustering. The key behind the Machine Learning model is to learn from data and build the model.[2]. When it receives new data, it predicts the output for new data. The Volume of data determines the accuracy of the predicted output. If the volume of data is high only the model predicts the output as more accurate.
-
To collect periodically bore-well water from our surroundings
-
To calculate water quality parameters TDS, pH, COD, BOD, F, Ca, and Mg hardness.
-
To find out the Water Quality Index by taking an average of all the parameters.
-
Based on the Gini Index Value, we construct a Decision Tree using its algorithm in Python.
-
Decision Tree and K-Nearest Neighbor algorithms are used to find out the model performance.
-
We developed models using Google Colab that use the K-nearest neighbor and Decision Tree algorithms to predict water quality in real time.
Random water samples are collected from several areas around our village. We have collected water samples from various Educational Institutions like Schools, Colleges, and Universities. Nearly we collected 105 samples and physico-chemical characteristics of the collected water samples were examined and reported. The Processing flow of our methodology is shown in Fig. 1.
3.1 Using WAWQI Method to Calculate WQI
Step 1: To calculate the value of various physico-chemical water quality parameters.
Step 2: To find the Proportionality Constant K by using the formula K = (1/(1/ ∑𝑛)
Step 3: To calculate a quality rating for the nth parameter (q n)
where n = Number of parameters
using formula q n=100 { (v n-v io)/(s n-v io)}
v n = Estimated value of the nth parameter of the given sampling station.
v io= Ideal value of the n-th parameter in pure water
s n=Standard permissible value of the nth parameter.
Step 4: Calculate the unit weight for the nth parameter. Wn=(k/s n).
Step 5: Calculate Water Quality Index (WQI) using the formula, WQI = ((Σw n* q n )/Σ w n)
We calculated the water quality index of our samples stored in our own dataset. The pictorial form of this index refers to some samples is given in Fig. 2. Based on the level of the water quality index we reported the status of water is tabulated in Table 1and our selected samples water quality index is tabulated in Table 2.
| LEVEL OF WQI | STATUS OF WATER |
|---|---|
| from 0 to 25 | High |
| from 26 to 50 | Medium |
| from 51 to 75 | Low |
| from 76 to 100 | Very Low |
| greater than 100 | Unsuitable to use |
| S.NO | SAMPLE NO | WQI | STATUS |
|---|---|---|---|
| 1 | S1 | 45 | Medium |
| 2 | S2 | 25 | High |
| 3 | S3 | 49 | Medium |
| 4 | S4 | 24 | High |
| 5 | S5 | 39 | Medium |
| 6 | S6 | 55 | Low |
| 7 | S7 | 72 | Low |
| 8 | S8 | 59 | Low |
| 9 | S9 | 32 | Medium |
| 10 | S10 | 52 | Low |
| 11 | S11 | 77 | Very Low |
| 12 | S12 | 60 | Low |
| 13 | S13 | 34 | Medium |
| 14 | S14 | 44 | Medium |
| 15 | S15 | 21 | High |
| 16 | S16 | 78 | Very Low |
| 17 | S17 | 39 | Medium |
| 18 | S18 | 67 | Low |
| 19 | S19 | 20 | High |
| 20 | S20 | 45 | Medium |
The pictorial form of our samples is given in Fig. 2. These samples are categorized into training and test data of our selected three models.
4.1 Decision Tree
One of the quantile-supervised learning algorithms is the decision tree. This algorithm is mainly used for regression and classification tasks. The decision tree has different parts like branches, roots, internal nodes, and leaf nodes. By using the divide and conquer method only, Decision Tree searches to identify the root node within a tree. This process is continued until all node’s gini[4] values are calculated using the Entropy formula.
The salient features of decision tree algorithms
-
They require less effort for data preprocessing.
-
It doesn’t require any normalization of data.
-
Missing values in the dataset do not affect the construction of the Decision Tree.
-
We can easily get the result from the Decision Tree model.
When this occurs, it is known as data fragmentation, and it can often lead to overfitting. To reduce the complexity and prevent overfitting, pruning is usually employed; this is a process, which removes branches that split on features with low importance.
Pruning is the process of removing connections from a network to increase the speed of inference and reduce its storage size. Pruning of a network deletes the unneeded parameters from an overly parameterized network. The model’s fit can then be evaluated through the process of cross-validation. This classifier predicts more accurate results, particularly when the individual trees are uncorrelated.
To choose the best attribute at each node
We must select the best attribute in each node among multiple ways like information gain and Gini impurity. They help to evaluate the quality of each test condition and how well it will be able to classify samples into a class.
Entropy and Information Gain
Entropy is used to measure the uncertainty of data. It is an essential metric that helps to evaluate the quality of a model and its ability to make accurate predictions. Here we used this entropy to determine the best split at each node. By using entropy only, we can build more robust and accurate models. Information gain is related to Entropy. It measures the impurity of the sample values. It is defined by the following formula [7]
Entropy values lie between 0 and 1. The entropy value is zero when all samples in the data set, S, belong to the same class. If half of the samples are classified under one class and the other half of the samples are in another class, then the entropy value is 1. To select the best feature to split on and find the optimal decision tree, the attribute with the smallest amount of entropy should be used. The difference in entropy before and after a split on a given attribute is represented by Information gain. The attribute that has the highest information gain will produce the best split as it is doing the best job at classifying the training data according to its target classification.
4.2 K-means clustering
Among several unsupervised machine learning algorithms, K-means clustering is one of the most effective ones. K-means clustering assigns data points to clusters based on which reference point is closest after constructing a centroid for the appropriate number of classes. Choosing the K value is the key point of the K-means algorithm. Here, we've covered a common technique for choosing K in the machine learning K-means algorithm.
K-Nearest Neighbor Algorithm steps
Step 1: Choose the number of clusters as K.
Step 2: Select random K points or centroids.
Step 3: Assign each data point to its closest centroid. It forms the predefined K clusters.
Step 4: Calculate a new centroid of each cluster.
Step 5: To take an average of samples from the same cluster.
Step-5: To reassign each data point to the new closest centroid of each cluster.
Step 6: If no new reassignment occurs, the model is ready. Else, go to step 4.
4.3 Naive Bayes
The Naive Bayes algorithm is based on the Bayes Theorem. It is also one of the simplest supervised learning algorithms. Naive Bayes classifier is a fast, accurate, and reliable algorithm. Naive Bayes classifiers have high accuracy and speed on large datasets. Prior and Posterior probabilities are used in this algorithm. Figure 3 shows the median value of water quality index of our samples using this algorithm. The steps used in the Naïve Bayes Algorithm are listed below.
Ø For the given class labels, calculate the prior probability.
Ø Apply the Bayes formula, and find the posterior probability.
Ø The given input belongs to the class which has a higher probability.
We applied three algorithms Naive Bayes, K-Nearest Neighbor, and Decision Tree for developing our classification model with our dataset as input. The WQI values of our samples were also calculated through the Models. We have utilized the Naive Bayes, Decision Tree, and K-Nearest Neighbor classifiers. We got three different accuracies (Naive Bayes-High, Decision Tree-Low) using these three classifiers shown in Table 3. From the result, we conclude that Naive Bayes has the highest accuracy at 95%, while the Decision Tree has the lowest accuracy at 91%. Figure 3 displays the performance of our applied models.
| S.NO | MODEL | ACCURACY ( % ) |
|---|---|---|
| 1 | Decision Tree | 91 |
| 2 | K-Nearest Neighbor | 92 |
| 3 | Naive Bayes | 95 |
The performance of machine learning techniques such as the Naïve Bayes, Decision Tree, and K-Nearest Neighbor Model to predict the water quality index of our surrounding educational institutions. The six important variables pH, TC, DO, BOD, Nitrate, and Temp for calculating the Water Quality Index were obtained from our dataset. We got the result by applying the above machine learning algorithms. We intimate the importance of water quality to the educational institutions of those who have bore water with low value. In the future, research will be carried out to build models that combine the proposed methods with deep learning approaches to improve efficiency.
Author Contribution
I wrote the code in Google Collab and analyzed the result (V.Queen Jemila)We collected water samples and found the parameters for calculating WQI (M.Dhanalakshmi, M.Amutha)
Data availability statement
• The datasets generated and/or analysed during the current study are not publicly available due to our own data but are available from the corresponding author on reasonable request.
- Sd A.M. Morrison, K. Coughlin, J.P. Shine, B.A. Coull, A.C. Rex Receiver operating characteristic curve analysis of beach water quality indicator variables Appl. Environ. Microbiol., 69 (2003), pp. 6405-6411,
- M. Najafzadeh, A. Ghaemi Prediction of the five-day biochemical oxygen demand and chemical oxygen demand in natural streams using machine learning methods
- Valentina-Andreea Călmuc 1*, Mădălina Călmuc 1, Maria Cătălina Ţopa1, ela Timofti 1, Cătălina Iticescu 1, Lucian P. Georgescu 1 various methods for calculating the water quality index
- 2 L D., Spiliotis, M., Ntislidou, C., Kagalou, I., Bobori, D., Tsiaoussi, V., Lazaridou, M., 2021. “One Out–All Out” Principle in the Water Framework Directive 2000—A New Approach with Fuzzy Method on an Example of Greek Lakes. Water (Switzerland) 13. https://doi.org/10.3390/w13131776.
- Mehedi Hassan1, *, Md. Mahedi Hassan2, Laboni Akter3, Md. Mushfiqur Rahman4, Sadika Zaman1, Khan Md. Hasib5, Nusrat Jahan6, Raisun Nasa Smrity2, Jerin Farhana7, M. Raihan1, Swarnali Mollick8 - Efficient Prediction of Water Quality Index (WQI) Using Machine Learning Algorithms
- Amir Hamzeh Haghiabi;Ali Heidar Nasrolahi;Abbas Parsaie Water quality prediction using machine learning methods
- T.Suryakanthi, Evaluating the Impact of GINI Index and Information Gain on Classification using Decision Tree Classifier Algorithm, International Journal of Advanced Computer Science and Applications January 2020
- Abazi A.M.S., Durmishi B.H., Sallaku F.S., Cadraku H.S., Fetoshi O.B., Ymeri P.H., Bytyci P.S. Assessment of water quality of sitnica river by using water quality index (WQI) RASAYAN J. Chem. 2020;13(1):146–159. [Google Scholar]
- Badan Pengendalian Lingkungan Hidup Kabupaten Bandung . Pemerintah Kabupaten Bandung Provinsi Jawa Barat: Bandung; Indonesia: 2015. Laporan Status Lingkungan Hidup Daerah Kabupaten Bandung.
- Cude C.G. Oregon water quality index: a tool for evaluating water quality management effectiveness. J. Am. Water Resour. Assoc. 2001;37(1):125
- Darvishi G., Kootenaei F.G., Ramezani M., Lotfi E., Asghamia H. Comparative investigation of river water quality by OWQI, NSFWQI, and wilcox indexes (case study: the Talar River – Iran) Arch. Environ. Protect. 2016;42(1):41–48]
- Davies J. Application and test of the Canadian water quality index for assessing changes in water quality in lakes and rivers of central north America. Lake Reservoir Manag. 2016;22(4):308–320]
No competing interests reported.