1

WATER QUALITY INDEX ESTIMATION USING

MACHINE LEARNING ALGORITHMS

Students:
Hussain Muhammad Tariq 20BNCIV-1077
Syed Rasan Ali 20BNCIV-1068
Hamid Khan 20BNCIV-1093
Umar Asghar 20BNCIV-1097
Supervisor:
Assistant Professor Engr. Abdus Salam

2
SCHEME OF PRESENTATION
INTRODUCTION

PROBLEM STATEMENT

OBJECTIVES OF RESEARCH

METHODOLOGY

RESULTS AND DISCUSSIONS

CONCLUSION

WORK SCHEDULE
Fig. Water Cycle
REFERENCES Source:https://web.facebook.com/tot.rootss/posts/-
3
the-water-cycle/349328976514634/?_rdc=1&_rdr
INTRODUCTION

 Water quality is vital for human well-being, ecosystems, and economic

development.

 80% of diseases in agricultural nations are water-borne, causing millions of deaths

and illnesses (Yogalakshmi and Mahalakshmi 2021).

 Surface water contamination is due to urbanization, industry, and agriculture.

 Accurate water quality assessment is necessary for environmental health and

stakeholders for taking suitable actions to overcome these issues.

4
INTRODUCTION

 Proposed research focuses on estimating the Water Quality Index by using

machine learning algorithms like:

 SVR (Support Vector Regression)
 Random Forest
 AdaBoost
 Decision Tree
 K-NN (K- Nearest Neighbor)

 Water quality estimation is helpful for better decision-making, early pollution

detection, and sustainable management.

5
INTRODUCTION

WATER QUALITY INDEX (WQI)?

 Water quality index (WQI) is a single number that represents overall quality of

water, calculating by taking into account the concentration of various water quality
parameters i.e. Ca (Calcium), Mg (Magnesium), Na (Sodium), HCO3
(Bicarbonate), Cl (Chloride), SO4 (Sulfate), Dissolved Solids by Evaporation,
Electrical Conductivity at 25°C , pH, Temperature, and Suspended Solids in PPM
(Parts Per Million).

 WQI is calculated by assigning weight to each of the available parameters by

using arithmetic method proposed by Brown et. al. in 1972. 6

PROBLEM STATEMENT

 Water Quality Index Estimation using Machine Learning Algorithm.

 Conventional WQI calculation methods often rely on manually intensive and time-

consuming data processing, making real-time monitoring and early detection of water
quality issues.

 By developing a machine learning-based WQI estimation system, this research aims to

address the limitations of traditional methods and improve the effectiveness of water
quality monitoring.

7
OBJECTIVES OF RESEARCH
 To develop a machine learning-based WQI estimation system for improved surface

water quality assessment.

 To collect the data from the Indus river and process it.

 To design and train machine learning models for WQI estimation.

 To evaluate model performance through cross-validation and comparisons.

 To provide recommendations for integrating the machine learning-based WQI system

into water quality monitoring practices.

8
 METHODOLOGY
STUDY AREA:

STUDY AREA

 The area selected for this Khairabad

study is the Indus river basin

at Khairabad station,
Nowshera, KP, Pakistan.

 The coordinates are 33.9° N,

72.22° E.

9
METHODOLOGY

10
METHODOLOGY

WQI CALCULATION:
As per (Torky, Bakhiet et al. 2023), the WQI was calculated using Arithmetic
weightage method by following steps

Where Sn is the standard value for each variable of water elements and K is a
constant. The weighted value of each element can be calculated as in below
equation

The Quality Impact value for each element in the water dataset can be calculated as
in below equation.

Finally, the water quality index can be calculated as in following equation

11
 METHODOLOGY
WQI CALCULATION
Standard
S.No Parameters Units 1/Sn K value Weightage
Values (Sn)
1 Ca (Calcium) mg/L 75 0.01333 4.591 0.061
2 Mg (Magnesium) mg/L 50 0.02000 4.591 0.092
3 Na (Sodium) mg/L 200 0.00500 4.591 0.023
4 (Bicarbonate) mg/L 120 0.00833 4.591 0.038
5 Cl (Chloride) mg/L 250 0.00400 4.591 0.018
6 (Sulfate) mg/L 400 0.00250 4.591 0.011
7 Dissolved Solids by Evaporation mg/L 600 0.00167 4.591 0.008
8 Electrical Conductivity at 25°C µS/cm 300 0.00333 4.591 0.015
9 pH - 8.5 0.11765 4.591 0.540
10 Temp (°C ) (Temperature) °C 25 0.04000 4.591 0.184
Suspended Solids in PPM (Parts
11 PPM 500 0.00200 4.591 0.009
Per Million)
Weightage
Sum of 1/Sn = 0.218 1
Sum
12
METHODOLOGY

WQI CALCULATION
 WQI was calculated using above formulae, for 11 parameters listed in previous
slide for 25 years in excel sheet (..\..\DATA\Study\WQI weightage.xlsx).
 Then WQI column was added to the original data, for the evaluation of machine
learning algorithms. (..\..\DATA\Study\Indus river at khairabad updated.xlsx).

13
Results and Discussions

 Codes for all five (i.e. AdaBoost, Decision Tree, K-NN (K- Nearest Neighbor),
Random Forest, SVM (Support Vector Model/Machine)) machine learning
models was generated.
 The performance of each model is evaluated and it’s scatter and swarm plots we
plotted which will be discussed in coming slides.

14
Results and Discussions
Performance Summary

Training Training Testing Testing

Algorithm
Set MSE Set R² Set MSE Set R²
Support Vector
1.02 0.93 4.06 0.88
Regression (SVR)
Random Forest 1.30 0.94 2.41 0.82
AdaBoost 0.19 0.99 0.81 0.90
Decision Tree 3.24 0.87 1.88 0.78
K-Nearest Neighbours
2.92 0.78 9.67 0.72
(KNN)

15
 Results and Discussions
Comparison Analysis
1.2

0.99
1
0.93 0.94
0.88 0.9 0.87
0.82
0.8 0.78 0.78
0.72

0.6

Training Set R²
0.4
Testing Set R²

0.2

0
Support Vector Random Forest AdaBoost Decision Tree K-Nearest
Regression (SVR) Neighbours
(KNN) 16
 Results and Discussions
Comparison Analysis

Training Set MSE

3.5
3.24
3 2.92

2.5

1.5 1.3
1.02
1

0.5
0.19
0
Support Vector Re- Random Forest AdaBoost Decision Tree K-Nearest Neighbours
gression (SVR) (KNN)
17
 Results and Discussions
Comparison Analysis

Testing Set MSE

12

10 9.67

4.06
4

2.41
2
1.88
0.81
0
Support Vector Random Forest AdaBoost Decision Tree K-Nearest Neighbours
Regression (SVR) (KNN)
18
 Results and Discussions
Support Vector Regression (SVR) Evaluation

Training and Testing Support Vector Regressor Model 19

 Results and Discussions
Support Vector Regression (SVR) Evaluation

Training and Testing Support Vector Regressor Model 20

 Results and Discussions
Random Forest Evaluation

Training and Testing of Random Forest Model 21

 Results and Discussions
Random Forest Evaluation

Training and Testing of Random Forest Model 22

 Results and Discussions
AdaBoost Evaluation

Training and Testing of AdaBoost Model 23

 Results and Discussions
AdaBoost Evaluation

Training and Testing of AdaBoost Model 24

 Results and Discussions
Decision Tree Regressor Evaluation

Training and Testing of Decision Tree Model

25
 Results and Discussions
Decision Tree Regressor Evaluation

Training and Testing of Decision Tree Model 26

 Results and Discussions
K-Nearest Neighbours (KNN) Evaluation

Training and Testing of K-NN Model 27

 Results and Discussions
K-Nearest Neighbours (KNN) Evaluation

Training and Testing of K-NN Model 28

STUDYConclusion
AREA:

 This study utilized different machine learning algorithms like SVR, Random Forest,

AdaBoost, Decision Tree and K-NN.

 Among which the best performer was AdaBoost with R² score: 0.99 (training), 0.90

(testing) which suggest it’s strong predictive capabilities and reliable tool for WQI,
while SVR was the second best with 0.88 R² score in testing.

29
 WORK
STUDY AREA: SCHEDULE

2023 2024
Activity
October November December January Feb …. …… July

Data Collection

Literature
Review
Learning
Models
Results

Thesis Write up

Submission 30
 REFERENCES
STUDY AREA:

 Yogalakshmi, S. and A. Mahalakshmi (2021). "Efficient water quality prediction for

Indian rivers using machine learning." Asian Journal of Applied Science and
Technology 5(1): 100-109.

31
32