1
Predictive Modeling of Water Quality and Sewage Systems: A Comparative Analysis and
Economic Impact Assessment Using Machine Learning
Md Fakhrul Islam Sumon
, Arifur Rahman
, Pravakar Debnath
, MD Rashed Mohaimin
4
, Mitu
Karmakar
, MD Azam Khan
and Hossain Mohammad Dalim
12567
School of Business, International American University, Los Angeles, California, USA
3
School of Business, Westcliff University Irvine, California, USA
4
MBA in Business Analytics, Gannon University, Erie, PA, USA
Corresponding Author:
Md Fakhrul Islam Sumon, E-mail: sumonf836@gmail.com
Abstract
Maintaining high water quality and effective sewage systems is imperative for the USA's
environmental sustainability and public health. Present issues related to water quality management
and effectively working sewage systems in the USA are multi-dimensional. Aging infrastructure,
lack of treatment facilities, and the absence of real-time monitoring systems are major
impediments to maintaining water quality. This study aimed at resolving the pressing matters
associated with water quality and sewage system efficiency through a multi-faceted approach. The
research project strived to ascertain the relationship between sewage system efficiency and overall
water quality. Besides, the present study endeavored to utilize machine learning techniques to
develop forecasts of future trends in water quality. The datasets were gathered from as many
reliable governmental databases as possible and environmental monitoring agencies to ensure
robust and correct analysis. Among other sources, the national water quality databases include
USGS, EPA, and EEA. These sources provided comprehensive data on a wide range of water
quality parameters, such as pH levels, dissolved oxygen (DO), biological oxygen demand (BOD),
chemical oxygen demand (COD), turbidity, nitrate and phosphate concentrations, and the presence
of heavy metals like lead, mercury, and cadmium. In this research project, three evidence-based
algorithms were selected, notably, Linear Regression, Random Forest, and XG-Boost are three
algorithms of machine learning that have been selected for performing predictive modeling.
Several performance metrics of the classes were performed for the stringent assessment of the
performance of Recall, Accuracy, Precision, and F1 Score machine learning models. The
performance of the Random Forest Classifier achieved an outstanding accuracy as compared to
other models. The findings of this study have great implications for water quality management in
the USA, especially concerning how predictive models could be leveraged further to advance
monitoring and intervention strategies. This provides the possibility to combine machine learning
algorithms in water quality management agencies that go beyond regular reactive approaches to
proactive data-driven strategies.
Keywords: Predictive Modelling; Water Quality; Sewage System; Economic Impact; Machine
Learning; Random Forest
2
I.
Introduction
Motivations and Background
Maintaining high water quality and effective sewage systems is paramount for the USA's
environmental sustainability and public health. Clean water is required not only for drinking
purposes but also for agriculture, industry, and ecosystem support. Equally important efficient
sewage systems whereby contaminants do not enter the water bodies and aquatic life is well
protected, ensuring safe water for human usage (Singh et. al., 2024). The completely unexpected
pace of urbanization and industrialization, even climate changes, has worsened the challenge of
managing water quality and sewage systems in the USA. These factors add pollutants to
freshwater bodies and overload water sewage infrastructure hence are inefficient and may be
disastrous (Talukdar et al., 2024).
According to Akhlaq et al. (2024), current problems pertinent to water quality management
and effectively working sewage systems are multi-dimensional. Aging infrastructure, lack of
treatment facilities, and the absence of real-time monitoring systems are major impediments to
maintaining water quality. There is also the added prevalence of substances like heavy metals,
pharmaceuticals, and microplastics that most methods of treatment cannot effectively deal with.
Ejaz et al. (2024), indicated that sewage systems within many regions are also ill-equipped to deal
with the raised volumes produced by booming populations; this mostly results in discharge into
the environment with no or partial treatment. These are challenges that call for drastic, innovative
solutions that will help manage water quality more efficiently and the processes involved in
sewage treatment.
Ahmed et al. (2024), argued that the economic and health repercussions of poor water
quality and insufficient sewage systems are significant. Contaminated water represents a potential
source for the spread of waterborne diseases such as cholera, typhoid, and hepatitis, causing serious
public health hazards, especially in low-income communities. Furthermore, Ameer et al. (2024),
asserted that poor water quality reduces agriculture and fisheries, hence creating food-insecure
communities where people lose their sources of livelihood. Economically, health costs, loss of
man-hours, and environmental clean-up of poor water management are very high. Therefore,
investment in efficient sewage systems and water quality management is not just a question of
public health but an economic one.
Objective
This study aims to resolve the pressing matters associated with water quality and sewage
system efficiency through a multi-faceted approach. First, the research project will strive to
ascertain the relationship between sewage system efficiency and overall water quality. The various
indicators in which these sciences are interrelated, such as levels of pollutants, efficacy of
treatment, and sewage system capacity, will be studied. The second objective is to compare the
governance of water quality in different regions to draw upon best practices and deficiencies in
these areas, mainly within an urban and rural setting. The third critical objective will involve the
investigation of the economic implications of sewage systems that are inadequate and have poor
water quality. Understanding the economic impacts of such consequences provides policymakers
with information to help prioritize investments in future water and sewage infrastructure. Lastly,
this present study endeavors to utilize machine learning techniques to develop forecasts of future
trends in water quality.
3
II.
Literature Review
Water Quality and Sewerage Systems
As per Asadollah et al. (2021), the maintenance of water quality is governed by some major
parameters and standards, which are used as yardsticks for safety and usability in drinking
applications or other uses in agriculture or industry in America. Key parameters monitored include
pH, dissolved oxygen, turbidity, BOD, and the presence of contaminants such as heavy metals and
pathogens within national and internationally accepted standards. Based on this, organizations like
WHO and EPA have set guidelines that stipulate permissible limits for such parameters, which
make the water safe for consumption and use.
With these standards, nonetheless, sewage systems in the USA are faced with various
issues conflicting with water quality. Among the most common problems that plague sewage
systems include aged infrastructure, incomplete treatment facilities, and poor disposal of industrial
and household waste. These frequently end up in discharging untreated or partially treated sewage
into natural water bodies, thereby contaminating freshwater. The inefficiency of sewage systems
is one of the major contributory causes affecting water quality, especially in the case of urban
areas, where wastewater production exceeds the capacities of existing facilities (Miller et al.,
2024). Many studies have been conducted on different modifications in sewage treatment
techniques, including advanced filtration technologies, bioremediation techniques, and
optimization in sewage network designs. These studies bring out the need for integrated solutions
with multi-faceted, multifactorial, technical, and policy-related challenges in water quality
management (Omeka, 2024).
The Economic Impact of Water Quality
The economic ramifications of poor water quality and sub-standard sewage systems are
profound and far-reaching. Poor water quality induces the spread of waterborne diseases, increases
healthcare costs, and lessens workforce productivity. The economic load is heavier on low-income
communities that have no access to clean water and efficient sewage systems, often resulting in
socioeconomic disparities in the long term. Research has documented that communities affected
by poor water quality endure increased medical expenses, lower agricultural yields, and reduced
property values which feed a self-reinforcing cycle of poverty and economic instability.
Görenekli & Gülbağ (2024), posited that case studies from various parts of the world have
been indicating large economic burdens of water pollution. For instance, the research on the
Ganges River in India showed that contamination of this vital watercourse has serious health
consequences and is extremely expensive regarding healthcare, tourism, and fisheries. To the same
extent, research on the Flint water crisis in the United States has demonstrated several long-term
economic consequences observable in the community, which vary from lower house property
values to higher public health expenditures. These examples illustrate a dire need for investments
in water quality improvement and sewage system upgrades that could help reduce economic losses
(Mukonza, 2024).
Machine Learning in Environmental Management
According to Van Nguyen et al. (2022), in the recent past, machine learning (ML) has
emerged as an instrumental tool in environmental management, specifically in forecasting and
mitigating the impacts of pollution. The algorithms of machine learning can analyze huge data to
predict patterns and trends which may not emerge conventionally by statistical methods. Other
applications of ML in environmental sciences include air quality indices prediction, modeling
scenarios of climate change impacts, and assessment of trends in water quality. Predictive
4
capabilities make for proactive environmental management such that interventions can be taken in
time, which may prevent or reduce pollution.
Zhu et al. (2022), articulated that application of machine learning in water quality
prediction has already witnessed several accomplishments. For instance, research has
demonstrated that ML models can predict the concentration of certain contaminants, such as
nitrates, and phosphates-continuing and vital water quality indicators. These models have been
applied in a decision-support context for the management of water resources to enable public
authorities to take precautions guaranteed to safeguard human health and protect the natural
environment. Despite these successes, there are limits to how machine learning can be applied in
this domain. There are several conditions when performance for the ML models depends
extensively on the quality and amount of the input data. Water quality data are scarce or
inconsistent in many parts of the world. Generalizing it across geographical and socio-economic
contexts may be problematic since the environmental systems are very complex. Nevertheless, the
role of machine learning could prove highly influential in changing the way water quality
management is done, especially concerning improving collection and processing technologies.
III.
Data Collection and Preprocessing
The foundation of this study lies in the extensive collection and analysis of datasets
associated with water quality and sewage system efficiency. The datasets were gathered from as
many reliable governmental databases as possible and environmental monitoring agencies to
ensure robust and correct analysis. Among other sources included the national water quality
databases include USGS, EPA, and EEA. These sources provided comprehensive data on a wide
range of water quality parameters, such as pH levels, dissolved oxygen (DO), biological oxygen
demand (BOD), chemical oxygen demand (COD), turbidity, nitrate and phosphate concentrations,
and the presence of heavy metals like lead, mercury, and cadmium. These range from critical
indications of water quality to the water div's health and its suitability for use by humans, aquatic
life, and agriculture.
Data-Preprocessing
Step 1-Datetime Handling:
First, 'Sampling Date' was converted into a proper date-time format
using pd.to_datetime(), where coercion of parsing errors is enabled. This procedure enabled a wide
range of data manipulations and extractions that can be performed efficiently later in the process.
Step 2-Encoding of Categorical Variables:
Label encoding was performed over the categorical
column 'State of Sewage System'. This protocol transformed the text categories into numerical
values, which are more suitable for machine learning algorithms.
Step 3-Handling Missing Values:
df. isnull().sum() code checked for missing values in the
dataset, indicating gaps that might not have been originally included. For continuous numerical
columns like 'Nitrogen (mg/L)' and 'Phosphorus (mg/L)', missing values were imputed using the
mean. In the case of date-time data, the mode is used to fill in the missing dates so that there will
not be any gaps in the dataset for analysis.
Step 4-Feature Engineering:
New features 'Year', 'Month', and 'Day' were extracted from
'Sampling Date' to capture the temporal patterns in data. This helped in improving the model
performance by leveraging time-based trends. After feature extraction, the original column
'Sampling Date' will be dropped as it's not needed anymore in its earlier form.
5
Step 5-Scaling Numerical Features:
StandardScaler() code standardized the numerical features,
including geographical coordinates and nutrient levels. This normalizes the feature values into a
scale that is similar, which may be important for algorithms sensitive to the magnitude of features.
Step 6-Data Split
: The last step divided the dataset into the necessary training and testing subsets
by applying the 80-20 split using train_test_split with test_size=0.2. For the given problem, the
target variable was the 'State of Sewage System', while the rest of the features were the predictor
variables. Setting a random state ensures the reproducibility of the split.
Exploratory Data Analysis (EDA)
Figure 1: Showcases the Distribution of Nitrogen and Phosphorous
The above graphs outline two of the most important water quality parameters, Nitrogen in mg/L
on the left and Phosphorus in mg/L on the right. The histograms, together with kernel density
estimates, are reasonably symmetrical and close to normally distributed, though not without
obvious multimodal happenstances. The Nitrogen levels make a cluster around an average of 0
after scaling, probably standardization; the greatest part of the data lies between -1.5 and +1.5
along the scaled axis-data was transformed to have a mean close to zero. Also, in that respect, the
spread and center of the Phosphorus levels are similar, which suggests that both features were
normalized similarly. This is a relatively even distribution with no extreme peaks or troughs,
suggesting that the dataset is considerably well-balanced with the least skewness feature good for
machine learning models since such a distribution likely means no serious outliers or biases in
those variables. Tiny fluctuations of frequency could suggest that there is some natural variation
in environmental measures but do not indicate serious imbalances or abnormalities.
6
Figure 2: Depicts the Correlation Heatmap of Various Features
Above is the correlation heatmap showing various feature relations-geographical
coordinates, water quality parameters, sewage system state, and temporal components such as
Year, Month, and Day. Out of these, the 'State of Sewage System' is very poorly correlated with
Nitrogen - 0.01 and Phosphorus - 0.00, which means the effective factor of sewage systems within
this dataset does not linearly affect these nutrient levels. The geographical factors such as Latitude
and longitude, along with temporal features such as Year, Month, and Day, get less than minimal
correlation from the water quality parameters and sewage system efficiency. No strong variable
correlations existed; hence, these features will be almost independent and perhaps require
extensive, complex nonlinear modeling approaches to find the underlying pattern in the data. This
independence also would mean that no single feature is dominant in the dataset, hence a more
balanced input to any machine learning model.
7
Figure 3:Displays the Nitrogen & Phosphorous Levels by State of the Sewage System.
The box plots above compare nitrogen and phosphorus levels across three states, 1, and 2
of sewage systems. In both nutrients, the patterns of distribution are similar across all three states,
each with median values around 0 mg/L and ranging from approximately -1.75 to +1.75 mg/L.
There is a slight trend of increase in the dispersion or box size for both nitrogen and phosphorus
levels as the state number increases from 0 to 2, but it is minimal. The symmetrical distribution of
values around the median would indicate that in all states, normal distribution patterns are reflected
by outliers shown by whiskers extending similarly in both positive and negative directions. Such
consistency among states shows that the nutrient levels of the sewage system are relatively stable
regardless of whether it is operational or not in operational.
8
Figure 4:Visualizes Monthly Trend of Nitrogen and Phosphorous Levels.
The time series plot above shows the monthly trend of nitrogen and phosphorus levels,
ranging from 2012 to 2024. Both nutrients have similar oscillating patterns around 0 mg/L. The
data indicates high-frequency fluctuations in both nutrients, generally within the range of -0.25 to
0.25 mg/L. Notable features include the strong peak in nitrogen to approximately 1.0 mg/L and
the sudden drop in phosphorus to around -0.5 mg/L toward the end of this time series. The shaded
areas around each line represent confidence intervals or uncertainty ranges and show a relatively
consistent variance over this monitoring period. Both nutrients are on the same trend of seasonality
or even cyclicality; no high long-term upward or downward trend until those anomalous readings
at the end of the series.
IV.
Methodology
Feature Engineering and Selection
Feature engineering and selection are some of the most critical stages in the creation of
any machine learning model, especially when dealing with environmental data. Therefore, diverse
different techniques were used in the project to extract and engineer useful features from the raw
data. In particular, we decomposed temporal data from 'Sampling Date' into separate features like
'Year', 'Month', and 'Day' to capture seasonal patterns that may influence water quality. Categorical
variables were represented by the 'State of Sewage System', pre-processed into a numerical
encoding using label encoding. The reason for doing this was to convert the textual data into a
machine-readable format. Feature scaling was applied to numerical variables such as 'Nitrogen
(mg/L)' and 'Phosphorus (mg/L)'. This is a process that scales those variables within a standard
range, hence improving model convergence during the training process. Therefore, only those
statistical methods, such as correlation analysis, were applied for the selection of the most
predictive features, taking into consideration variables that show low multicollinearity to avoid
redundancy and overfitting. The aim was to retain those features that contribute substantially to
the target variable 'State of Sewage System', ensuring a balanced model with both accuracy and
interpretability.
Model Selection and Justification
In this research project, three evidence-based algorithms were selected, notably, Linear
Regression, Random Forest, and XG-Boost are three algorithms of machine learning that have
been selected for performing predictive modeling.
Linear Regression
was chosen because it is very
simple and efficient at capturing the linear relationship of independent variables with the target.
Therefore, this may act as a baseline model to understand the direct influence of features on sewage
9
system efficiency.
Random Forest,
an ensemble method based on decision trees, was adopted
because it can provide a complex nonlinear interaction without severe over-fitting via
bootstrapping and randomness in features. It is efficient in capturing intricate interactions between
features and gives feature importance scores, which will be useful in further feature selection. On
the other hand,
XG-Boost
was chosen for its excellent performance against large datasets with high
dimensionality. It combines the strengths of gradient boosting with regularization techniques;
hence, being highly effective at optimizing accuracy with lesser overfitting. XG-Boost is
acknowledged to be one of the most efficient and scalable algorithms in data science competitions.
Hence, it is suitable for this project: an accurate prediction of water quality trends.
Training and Testing Framework
In this research project, the dataset has been divided into an 80-20 split to ensure that
the model captures 80% of the data to train on and is tested on 20%. This protocol helped in
assessing the generalization capability of the model. To further increase the robustness in
evaluating the model, k-fold cross-validation was performed with k=5. It implies splitting the
training data into five folds, training the model sequentially on four folds while validating on the
fifth, through all possible rotations. Cross-validation helps prevent the problem of overfitting by
ensuring that the performance of a model is consistent across different subsets of the data. Besides,
hyperparameter tuning is also done through a grid search for better performance regimes of the
model parameters. Performance metrics evaluated are MAE, RMSE, and R-squared were used to
assess model accuracy and robustness.
Hyperparameter Tuning
Optimizing model performance involves tuning the hyperparameters, which control the
learning process and behavior of machine learning algorithms. In this study, two major approaches
were used for hyperparameter tuning, namely: Grid Search and Random Search. In Grid Search,
the approach considers a pre-defined set of combinations of hyperparameters to explore
systematically and retrieve the best parameters that maximize model performance. In contrast,
Random Search selects random combinations of hyperparameters within specified ranges. The
latter approach is much quicker for large parameter spaces compared to Grid Search and therefore
best suited to efficiently explore large parameter spaces. It was especially helpful at the beginning
of the experimentation for quickly determining promising bounds of hyperparameters for further
fine-tuning. Using Grid Search when precision is important and Random Search when speed is
important yields a good balance in optimizing model performance while avoiding extreme
computational costs.
Performance Evaluation Metrics
Several performance metrics of the classes were performed for the stringent assessment
of the performance of Recall, Accuracy, Precision, and F1 Score machine learning models. These
metrics gave a complete understanding of the effectiveness that models may have, especially in
cases where classes are highly imbalanced, or the costs of false positives and false negatives are
very different. In the baseline testing performance of selected models Random Forest and XG-
Boost-their evaluation metrics are compared to those of some baseline model, such as Logistic
Regression or a Decision Tree classifier. This baseline provides a reference to allow qualification
of the added value when using more sophisticated algorithms. Baseline models are characterized
by decent accuracy, for example, but they may be substantially worse about recall and precision,
especially events that occur less often such as severe sewage problems.
10
V.
Results
Descriptive Analysis
Performance Metric
Random Forest
XG-Boost
Logistic Regression
Accuracy
99.60%
82.40%
50.29%
Precision [class 0]
0.99
0.77
0.50
Precision [class 1]
1.00
0.91
0.00
Precision [Class 2]
1.00
0.96
0.00
Recall [class 0]
1.00
0.97
1.00
Recall [class 1]
0.99
0.73
0.00
Recall [Class 2]
0.99
0.58
0.00
F1-Score [Class 0]
0.99
0.86
0.67
F1-Score [Class 1]
1.00
0.81
0.00
F1_Score [Class 2]
1.00
0.72
0.00
The Table above displays the performance results comparing three models: Random
Forest, XG-Boost, and Logistic Regression. The best classification performance, according to the
above table, is from the Random Forest, which yields an accuracy of 99.60%. Compared to other
models, it depicts powerful performance among all metrics, including perfect or near-perfect
precision, recall, and F1-scores belonging to all classes. The XG-Boost model follows, presenting
an accuracy of 82.40% only. The performance of XG-Boost for the two classes is significantly
lower than for the other two methods, with significant differences in recall and F1-score measures.
Logistic Regression, in turn, performs considerably worse, yielding an accuracy of only 50.29%,
completely misclassifying classes 1 and 2, while performing quite well for class 0. This finding
also confirms the robustness of Random Forest on this data set, while the performance of Logistic
Regression is comparatively poor in terms of multi-classification tasks.
Model Performance
A.
Logistic Regression
# Logistic Regression
log_reg
=
LogisticRegression(max_iter
=
1000, random_state
=
42)
log_reg
.
fit(X_train, y_train)
y_pred_log_reg
=
log_reg
.
predict(X_test)
# Evaluation
print("Logistic Regression Results:")
print("Accuracy:", accuracy_score(y_test, y_pred_log_reg))
print("\nClassification Report:\n", classification_report(y_test,
y_pred_log_reg))
print("\nConfusion Matrix:\n", confusion_matrix(y_test, y_pred_log_reg))
Table 1: Portrays the logistic Regression Modelling
The code above performs binary classification using the Logistic Regression model. First,
the model is instantiated with a maximum iteration of 1000 and a random state for reproducibility.
Then it fits into X_train and y_train data using the fit() method and makes predictions on data
X_test. The code also includes an extensive evaluation section that prints several performance
metrics: the accuracy score of the model; the detailed classification report which, among others,
11
includes precision, recall, and F1-score; and finally, it also prints a confusion matrix. These are
enough to provide a comprehensive review of the model's performance in classifying test data.
Output:
Classification Report:
precision recall f1-score support
0 0.50 1.00 0.67 4031
1 0.00 0.00 0.00 2519
2 0.00 0.00 0.00 1466
accuracy 0.50 8016
macro avg 0.17 0.33 0.22 8016
weighted avg 0.25 0.50 0.34 8016
Table 2: Presents the Logistic Regression Classification Report
As showcased above, Logistic regression had an average performance of 50.3%. From the
classification report, serious issues can be identified: only class 0 examples are classified correctly;
it has a precision of 0.50 with a recall of 1.00, indicating that it predicts everything as class 0. This
dataset is imbalanced, with the following distribution: class 0 with 4,031 samples, class 1 with
2,519 samples, and class 2 with 1,466 samples. It is confirmed by very low metrics for the macro
average, an unweighted mean across classes, and weighted average, which refers to different
metrics weighted averages considering the class supports. The macro average F1-score of 0.22 and
weighted average F1-score of 0.33 lead us to believe that this model was average; important
ameliorations need to be performed.
B.
Random Forest
# Random Forest Classifier
rf_clf
=
RandomForestClassifier(n_estimators
=
100, random_state
=
42)
rf_clf
.
fit(X_train, y_train)
y_pred_rf
=
rf_clf
.
predict(X_test)
# Evaluation
print("\nRandom Forest Classifier Results:")
print("Accuracy:", accuracy_score(y_test, y_pred_rf))
print("\nClassification Report:\n", classification_report(y_test, y_pred_rf))
print("\nConfusion Matrix:\n", confusion_matrix(y_test, y_pred_rf))
Table 3: Depicts the Random Forest Modelling
The code snippet above creates a Random Forest Classifier, an ensemble learning method
that builds on generating multiple decision trees. An instance of the model is created with 100
estimators (the decision trees) and a state (for reproducibility) of 42. As seen previously with the
code for logistic regression, fit() is used to fit the model to some training X and y data and then
predict some test X data. The evaluation uses the same metrics as above: accuracy, classification
report, and confusion matrix.
12
Output:
Classification Report:
precision recall f1-score support
0 0.99 1.00 1.00 4031
1 1.00 0.99 1.00 2519
2 1.00 0.99 1.00 1466
accuracy 1.00 8016
macro avg 1.00 1.00 1.00 8016
weighted avg 1.00 1.00 1.00 8016
Table 4: Exhibits the Random Forest Classification Report
The performance of the Random Forest Classifier achieved an outstanding accuracy of
99.6%. It can also be observed that almost perfect classification among the classes is realized, 0,
1, and 2, with precision, recall, and F1-scores being exactly 1.00. Model performance for class 0
results in 4,031 samples being correctly classified with 0.99 precision and 1.00 recall, while classes
1 and 2, by convention, have 2,519 and 1,466 samples correspondingly and result in perfect
precision of 1.00 and almost perfect recalls of 0.99 each. Both the macro and weighted averages
are also 1.00 across all metrics, which further indicates balanced and superior performance across
class imbalances. This represents a dramatic improvement from the Logistic Regression results
and indicates that the Random Forest Classifier is much better suited for this particular
classification task.
C.
XG-Boost
# XGBoost Classifier
xgb_clf
=
XGBClassifier(use_label_encoder
=False
, eval_metric
=
'logloss',
random_state
=
42)
xgb_clf
.
fit(X_train, y_train)
y_pred_xgb
=
xgb_clf
.
predict(X_test)
# Evaluation
print("\nXGBoost Classifier Results:")
print("Accuracy:", accuracy_score(y_test, y_pred_xgb))
print("\nClassification Report:\n", classification_report(y_test,
y_pred_xgb))
print("\nConfusion Matrix:\n", confusion_matrix(y_test, y_pred_xgb))
Table 5: Portrays the XG-Boost Classifier Modelling
This code snippet above executes an XG-Boost Classifier, a powerful gradient-boosting model
renowned for its performance and speed. One prepares the model with the following parameters:
label_encoder as false to handle the labels directly, eval_metric with 'log loss' to evaluate the model
performance using logarithmic loss and random_state equal to 42 to make the experiment
reproducible. Similar to previous examples, it follows the same pattern: fitting the model on the
training data (X_train, y_train), making predictions on the test data (X_test), and keeping
consistency in the evaluation section by outputting the accuracy score, classification report, and
confusion matrix as standard performance assessment means for the model.
13
Output:
Classification Report:
precision recall f1-score support
0 0.77 0.97 0.86 4031
1 0.91 0.73 0.81 2519
2 0.96 0.58 0.72 1466
accuracy 0.82 8016
macro avg 0.88 0.76 0.80 8016
weighted avg 0.85 0.82 0.82 8016
Table 6: Showcases the XG-Boost Classification Report
The above table presents the results of the XG-Boost Classifier model. The model has
correctly predicted 82.39% of all instances within this dataset. The classification report includes
detailed information on performances for each class. Class 0 has high recall-97%-with 77%
precision, which assumes good performance in identifying true positives. Class 1 has a rather
balanced precision of 91% and recall of 73%, showing that for this class, there is a good trade-off
between true positives identified and false positives raised. Class 2 has a lower recall of 58% and
precision of 96%, which can be indicative of problems correctly identifying the instances of this
class. Overall, the model performs well in terms of accuracy and precision. Nevertheless,
concerning class 2, there is room for further improvement in its recall.
Feature Importance and Correlation Analysis
Comprehending the key drivers beneath water quality and sewage system efficiency is
crucial for developing an efficient predictive algorithm. It is against this background that the use
of feature importance scores considers models such as Random Forest and Gradient Boosting that
are inherently useful in providing insights on which variables most drive predictions by calculating
the importance of each feature in determining the model output. The most influencing features of
the given study are Nitrogen and Phosphorus concentration in mg/L, Geographical Location, and
Sampling Date. For example, in the Random Forest model, the highest ranking in importance was
given to the nutrient levels, making changes in the non-turbidity parameters be strong predictor of
water quality deterioration linked to sewage system inefficiency. The same conclusion is
confirmed by the Gradient Boosting model since it highlights nutrient pollution. Such insights are
highly useful in interventions to be given at appropriate targets, as such insights on the part of
environmental agencies can prioritize monitoring and managing based on the factors that have a
greater impact.
Apart from feature importance, we also analyzed the correlation to understand how
sewage system efficiency might relate to the different water quality parameters. Nutrient-level
variables, such as Nitrogen and Phosphorus, showed a positive correlation with poor sewage
systems in the correlation heatmap; thus, inefficient sewage systems lead to higher concentrations
of such pollutants. Geographical coordinates along with temporal features like Year, Month, and
Day, though having low correlation coefficients, did their job in capturing seasonal or locational
variation in water quality. This analysis shows the diverse facets of water pollution, both of
anthropogenic and natural nature that interact.
Economic Impact Assessment
The economic effects of poor water quality and unmanaged sewage systems run very
deep, impacting many aspects of life: from public health and agriculture to tourism and general
14
community well-being. Poor sewage management that leads to pollution of water bodies increases
the rates of waterborne disease, causing health care costs to leap. Such communities are bound to
experience the spread of diseases as a result of untreated or poorly treated water, which exposes
people to cholera and gastroenteritis. This increases the cost of medication, hence resulting in the
loss of productive hours because of sickness. Furthermore, the poor quality of water significantly
impacts agricultural activities through irrigation water contamination, reducing crop yields, and
increasing farming costs related to water treatment. This leads to financial loss for the farmers and
raises prices for the consumers, thus having an impact on the entire value chain of food.
Indeed, numerous studies done across the United States testify to the huge economic
impacts of failing water and sewage systems. For example, there was the Flint, Michigan, water
crisis, wherein quite poor treatment processes led to a leakage of lead into the city's drinking water
supply. This not only poisoned scores of residents, with the worst effects felt by children but
brought in a piece of long-term economic devastation. Lawsuits against the city, sharp declines in
property values, millions of dollars in damages, and healthcare costs: were some of the costly
results. Apart from the loss of civic trust, there was massive investment to be made in rebuilding
the water infrastructure and restructuring the community's faith in public services.
Another example is the Mississippi River Basin, which has been polluted with nutrients
due to inefficient sewage systems and runoff from fertilized agricultural fields. High levels of
nitrogen and phosphorus have stimulated the growth of a large "dead zone" in the Gulf of Mexico
where aquatic life cannot survive because of a lack of oxygen and where fishing and tourism
industries are seriously affected. Thus, economic damage to the said commercial fisheries' activity
in this region has been estimated in hundreds of millions of dollars annually since hypoxic
conditions and oxygen levels make it hard for marine life to live. This reduction in fish stock
affects local fishers and impacts the overall economy dependent on the supply chain of seafood.
In Florida, the incidences of harmful algal blooms have continued to torture the state,
with increasing agricultural runoff and sewage treatment further delving into exacerbating the
problem. These have economic consequences, as tourism-based economies are especially affected
when beach closures and health advisories are issued, leading to losses in hotel bookings,
recreational activities, and local businesses. According to one estimate, the 2018 red tide in Florida
cost the state approximately $130 million in lost tourism. Examples like these are the underpinning
reasons why investment is critically needed in modern sewage systems, along with the
management system of water quality that will reduce these economic impacts. The investment in
infrastructure not only will protect public health and the environment but also will give long-term
economic benefits by reducing these basic economic burdens from damages related to pollution.
The novelty of such a dual focus lies in the combination of environmental and economic outcomes
concerning the importance of efficient sewage systems for sustainable development.
VI.
Discussion
Implications for Water Quality Management
The findings of this study have great implications for water quality management, especially
concerning how predictive models could be leveraged further to advance monitoring and
intervention strategies. This provides the possibility to combine machine learning algorithms in
water quality management agencies that go beyond regular reactive approaches to proactive data-
driven strategies. Predictive models project potential water quality problems based on history and
thus allow timely interventions to prevent contamination events and optimize sewage network
operations. Such models have the potential to automatically identify sources of pollution, predict
15
environmental changes that affect water quality, and perform optimal resource allocation to
monitoring efforts. For instance, this is possible in embedding machine learning models at
established environmental monitoring systems where the detecting accuracy of such pollutants as
nitrogen and phosphorus levels shall enable policymakers to establish more stringent regulatory
measures. It is recommended that user-friendly interfaces should be developed for environmental
agencies so that they can flawlessly embed predictive analytics into their day-to-day operations.
Challenges and Limitations
Notwithstanding, several limitations and challenges should be addressed to maximize the
benefits of these models. One such critical issue is the dealing of environmental data, especially
sensitive information having a bearing on water sources that communities may depend on. Data
privacy and conformity to regulatory requirements are very much in order. Similarly, model
performance is heavily influenced by data quality and quantity. Poor practices in the collection of
data, such as inconsistent frequency in data, missing values, or limits to real-time data access, can
decrease the accuracy of the models leading to unreliable predictions. Another challenge is
interpretability for such complex models as Gradient Boosting and Random Forest, because some
predictions cannot intuitively be understood by stakeholders and, hence, may stand in the way of
decision-making. Besides, generalization raises another limitation across different regions with
different environmental conditions. A model that performs well in one geographical area might
not perform well in another, first, because of the different water quality parameters of each place,
and second, mainly because of the different pollution sources of each area.
Future Research Directions
Forging ahead, future research directions can concentrate on resolving these limitations
and challenges by expanding the diversity of datasets used for model training. The diversities of
data from various regions and climatic conditions could make the models robust and generalizable.
There is also the possibility to examine the development of real-time water quality monitoring
with IoT devices and satellite imagery for streams to make more accurate and dynamic predictions.
Research into hybrid models can also be explored, which allows a combination of the key features
of various machine learning methods that may prove particularly effective in achieving greater
predictive accuracy. The future looks brighter as evolving technology will introduce more
advanced and large-scale machine learning applications to improve water quality management,
enhancing the outcomes for public health and environmental sustainability.
VII.
Conclusion
This study aimed at resolving the pressing matters associated with water quality and
sewage system efficiency in the USA through a multi-faceted approach. The research project
strived to ascertain the relationship between sewage system efficiency and overall water quality in
the USA. Besides, the present study endeavored to utilize machine learning techniques to develop
forecasts of future trends in water quality. The datasets were gathered from as many reliable
governmental databases as possible and environmental monitoring agencies to ensure robust and
correct analysis. Among other sources included the national water quality databases include
USGS, EPA, and EEA. These sources provided comprehensive data on a wide range of water
quality parameters, such as pH levels, dissolved oxygen (DO), biological oxygen demand (BOD),
chemical oxygen demand (COD), turbidity, nitrate and phosphate concentrations, and the presence
of heavy metals like lead, mercury, and cadmium. In this research project, three evidence-based
algorithms were selected, notably, Linear Regression, Random Forest, and XG-Boost are three
algorithms of machine learning that have been selected for performing predictive modeling.
16
Several performance metrics of the classes were performed for the stringent assessment of the
performance of Recall, Accuracy, Precision, and F1 Score machine learning models. The
performance of the Random Forest Classifier achieved an outstanding accuracy as compared to
other models. The findings of this study have great implications for water quality management,
especially concerning how predictive models could be leveraged further to advance monitoring
and intervention strategies. This provides the possibility to combine machine learning algorithms
in water quality management agencies that go beyond regular reactive approaches to proactive
data-driven strategies.
References
Ahmed, A. N., Othman, F. B., Afan, H. A., Ibrahim, R. K., Fai, C. M., Hossain, M. S., ... &
Elshafie, A. (2019). Machine learning methods for better water quality
prediction.
Journal of Hydrology
,
578
, 124084.
Akhlaq, M., Ellahi, A., Niaz, R., Khan, M., Sammen, S. S., & Scholz, M. (2024). Comparative
Analysis of Machine Learning Algorithms for Water Quality Prediction.
Tellus A:
Dynamic Meteorology and Oceanography
,
76
(1).
Al Mukaddim, A., Nasiruddin, M., & Hider, M. A. (2023). Blockchain Technology for Secure
and Transparent Supply Chain Management: A Pathway to Enhanced Trust and
Efficiency. International Journal of Advanced Engineering Technologies and
Innovations, 1(01), 419-446.
Al Mukaddim, A., Mohaimin, M. R., Hider, M. A., Karmakar, M., Nasiruddin, M., Alam, S., &
Anonna, F. R. (2024). Improving Rainfall Prediction Accuracy in the USA Using
Advanced Machine Learning Techniques. Journal of Environmental and Agricultural
Studies, 5(3), 23-34.
Alqahtani, A., Shah, M. I., Aldrees, A., & Javed, M. F. (2022). Comparative assessment of
individual and ensemble machine learning models for efficient analysis of river water
quality. Sustainability, 14(3), 1183.
Ameer, S., Shah, M. A., Khan, A., Song, H., Maple, C., Islam, S. U., & Asghar, M. N. (2019).
Comparative analysis of machine learning techniques for predicting air quality in
smart cities.
IEEE access
,
7
, 128325-128338.
Asadollah, S. B. H. S., Sharafati, A., Motta, D., & Yaseen, Z. M. (2021). River water quality
index prediction and uncertainty analysis: A comparative study of machine learning
models. Journal of Environmental Chemical Engineering, 9(1), 104599.
Buiya, M. R., Laskar, A. N., Islam, M. R., Sawalmeh, S. K. S., Roy, M. S. R. C., Roy, R. E. R.
S., & Sumsuzoha, M. (2024). Detecting IoT Cyberattacks: Advanced Machine
Learning Models for Enhanced Security in Network Traffic. Journal of Computer
Science and Technology Studies, 6(4), 142-152.
Debnath, P., Karmakar, M., & Sumon, M. F. I. (2024). AI in Public Policy: Enhancing Decision-
Making and Policy Formulation in the US Government. International Journal of
Advanced Engineering Technologies and Innovations, 2(1), 169-193.
Debnath, P., Karmakar, M., Khan, M. T., Khan, M. A., Al Sayeed, A., Rahman, A., & Sumon,
M. F. I. (2024). Seismic Activity Analysis in California: Patterns, Trends, and
Predictive Modeling. Journal of Computer Science and Technology Studies, 6(5), 50-
60.
17
Ejaz, U., Khan, S. M., Jehangir, S., Ahmad, Z., Abdullah, A., Iqbal, M., ... & Svenning, J. C.
(2024). Monitoring the Industrial waste polluted stream-Integrated analytics and
machine learning for water quality index assessment.
Journal of Cleaner
Production
,
450
, 141877.
Görenekli, K., & Gülbağ, A. (2024). Comparative Analysis of Machine Learning Techniques for
Water Consumption Prediction: A Case Study from Kocaeli
Province.
Sensors
,
24
(17), 5846.
Hasan, M. R., Islam, M. Z., Sumon, M. F. I., Osiujjaman, M., Debnath, P., & Pant, L. (2024).
Integrating Artificial Intelligence and Predictive Analytics in Supply Chain
Management to Minimize Carbon Footprint and Enhance Business Growth in the
USA. Journal of Business and Management Studies, 6(4), 195-212.
Islam, M. R., Nasiruddin, M., Karmakar, M., Akter, R., Khan, M. T., Sayeed, A. A., & Amin, A.
(2024). Leveraging Advanced Machine Learning Algorithms for Enhanced
Cyberattack Detection on US Business Networks. Journal of Business and
Management Studies, 6(5), 213-224.
Islam, M. R., Shawon, R. E. R., & Sumsuzoha, M. (2023). Personalized Marketing Strategies in
the US Retail Industry: Leveraging Machine Learning for Better Customer
Engagement. International Journal of Machine Learning Research in Cybersecurity
and Artificial Intelligence, 14(1), 750-774.
Karmakar, M., Debnath, P., & Khan, M. A. (2024). AI-Powered Solutions for Traffic
Management in US Cities: Reducing Congestion and Emissions. International Journal
of Advanced Engineering Technologies and Innovations, 2(1), 194-222.
Khan, M. A., Debnath, P., Al Sayeed, A., Sumon, M. F. I., Rahman, A., Khan, M. T., & Pant, L.
(2024). Explainable AI and Machine Learning Model for California House Price
Predictions: Intelligent Model for Homebuyers and Policymakers. Journal of Business
and Management Studies, 6(5), 73-84.
Kouadri, S., Elbeltagi, A., Islam, A. R. M. T., & Kateb, S. (2021). Performance of machine
learning methods in predicting water quality index based on irregular data set:
application on Illizi region (Algerian southeast).
Applied Water Science
,
11
(12), 190.
Miller, T., Durlik, I., Adrianna, K., Kisiel, A., Cembrowska-Lech, D., Spychalski, I., & Tuński,
T. (2023). Predictive Modeling of Urban Lake Water Quality Using Machine
Learning: A 20-Year Study.
Applied Sciences
,
13
(20), 11217.
Mukonza, S. S., & Chiang, J. L. (2023). Meta-Analysis of Satellite Observations for United
Nations Sustainable Development Goals: Exploring the Potential of Machine
Learning for Water Quality Monitoring.
Environments
,
10
(10), 170.
Nasiruddin, M., Al Mukaddim, A., & Hider, M. A. (2023). Optimizing Renewable Energy
Systems Using Artificial Intelligence: Enhancing Efficiency and Sustainability.
International Journal of Machine Learning Research in Cybersecurity and Artificial
Intelligence, 14(1), 846-881.
Omeka, M. E. (2024). Evaluation and prediction of irrigation water quality of an agricultural
district, SE Nigeria: an integrated heuristic GIS-based and machine learning
approach.
Environmental Science and Pollution Research
,
31
(41), 54178-54203.
Shil, S. K., Chowdhury, M. S. R., Tannier, N. R., Tarafder, M. T. R., Akter, R., Gurung, N., &
Sizan, M. M. H. (2024). Forecasting Electric Vehicle Adoption in the USA Using
Machine Learning Models. Journal of Computer Science and Technology Studies,
6(5), 61-74.
18
Singh, S., Das, A., & Sharma, P. (2024). Predictive modeling of water quality index (WQI)
classes in Indian rivers: Insights from the application of multiple Machine Learning
(ML) models on a decennial dataset.
Stochastic Environmental Research and Risk
Assessment
, 1-18.
Shawon, R. E. R., Rahman, A., Islam, M. R., Debnath, P., Sumon, M. F. I., Khan, M. A., &
Miah, M. N. I. (2024). AI-Driven Predictive Modeling of US Economic Trends:
Insights and Innovations. Journal of Humanities and Social Sciences Studies, 6(10),
01-15.
Sumon, M. F. I., Osiujjaman, M., Khan, M. A., Rahman, A., Uddin, M. K., Pant, L., & Debnath,
P. (2024). Environmental and Socio-Economic Impact Assessment of Renewable
Energy Using Machine Learning Models. Journal of Economics, Finance and
Accounting Studies, 6(5), 112-122.
Talukdar, S., Ahmed, S., Naikoo, M. W., Rahman, A., Mallik, S., Ningthoujam, S., ... &
Ramana, G. V. (2023). Predicting lake water quality index with sensitivity-uncertainty
analysis using deep learning algorithms.
Journal of Cleaner Production
,
406
, 136885.
Van Nguyen, L., Bui, D. T., & Seidu, R. (2022). Comparison of machine learning techniques for
condition assessment of sewer network.
IEEE Access
,
10
, 124238-124258.
Zeeshan, M. A. F., Sumsuzoha, M., Chowdhury, F. R., Buiya, M. R., Mohaimin, M. R., Pant, L.,
& Shawon, R. E. R. (2024). Artificial Intelligence in Socioeconomic Research:
Identifying Key Drivers of Unemployment Inequality in the US. Journal of
Economics, Finance and Accounting Studies, 6(5), 54-65.
Zhu, M., Wang, J., Yang, X., Zhang, Y., Zhang, L., Ren, H., ... & Ye, L. (2022). A review of the
application of machine learning in water quality evaluation.
Eco-Environment &
Health
,
1
(2), 107-116.
