Prediction of Ground Water Level using Machine Learning

Dr P Nageswara Rao1, N Yasaswini2, S Ayesha3, M Jayasri4, P Gowsiya5, Y Nagaswetha6

1
Professor, Dept. of CSE, Vemu Institute of Technology, P. Kothakota, AP, India
2,3,4,5,6
UG Student, Dept. of CSE, Vemu Institute of Technology, P. Kothakota, AP, India

Evaluation of machine learning models for

Abstract-Groundwater level prediction is a groundwater level prediction involves the use of
critical task in water resource management, various metrics to assess their accuracy, reliability,
environmental monitoring, and agricultural and generalization capabilities. Metrics such as
planning. Traditional methods rely on simplistic root mean square error (RMSE), mean absolute
statistical models and manual data analysis, which error (MAE), coefficient of determination (R-
often fail to capture the complex dynamics of squared), and Nash-Sutcliffe efficiency (NSE)
groundwater systems. In recent years, machine provide valuable insights into the predictive
learning (ML) techniques have emerged as performance of the models and their ability to
powerful tools for predicting groundwater levels
with higher accuracy and efficiency. This abstract capture the underlying dynamics of groundwater
presents a comprehensive overview of the systems.
application of machine learning algorithms in Real-world applications of machine learning in
groundwater level prediction. We discuss various groundwater level prediction encompass a wide
ML models, including regression-based methods, range of scenarios, including groundwater
neural networks, support vector machines, and management, drought forecasting, aquifer
ensemble learning techniques, highlighting their characterization, and early warning systems for
strengths and weaknesses in predicting groundwater-related hazards. Case studies and
groundwater levels. research efforts demonstrate the effectiveness of
Keywords: Data analytics, Numpy, Pandas, machine learning models in improving the
Logistic regression, Decision tree, SVM accuracy and timeliness of groundwater level
predictions, thereby facilitating informed
I.Introduction decision-making and sustainable management of
groundwater resources.
Groundwater level prediction is a crucial
component of water resource management,
environmental sustainability, and agricultural
planning. Traditional methods for predicting
groundwater levels often rely on simplistic
statistical models and manual data analysis
techniques. However, these methods struggle to
adequately capture the intricate dynamics inherent
in groundwater systems, leading to limitations in
accuracy and effectiveness. The emergence of
machine learning (ML) techniques has offered a
promising alternative, enabling more sophisticated
and data-driven approaches to groundwater level
prediction.

1|Page
II.Literature Review 3 S. Predictio The results 20
Choudh n of predict that 23
Sl.N Name Title Description Ye
ary and Ground increased
o. of the ar
M. Water Prosopis
Aurtho
Pundir Level in land cover
r(s)
Prosopis reduces the
1 Qazi Predictiv Regression 20 Juli flora groundwate
waqas e techniques 23 Ecosyste r level to 55
khan Modeling are used for m using meters
&Rashi of Water predicting Machine below
d Table water table Learning ground
Ahmed Depth, depth and level by
Drilling number of 2050 in
Duration, days. Tamil Nadu.
and Soil
4 S. Suggestio The study’s 20
Layer
Behera, n of findings 23
Classifica
D. Appropri will help
tion
Menon ate Crops farmers,
Using
Based on agricultural
Adaptive
Rainfall researchers,
Ensemble
and and
Learning
Undergro policymaker
und s make
2 LuLI & Estimatio Based on a 20 Water cropmanage
Aduwat n of three-layer 22 Analysis ment and
i Sali Ground and five water
Water input resource
Level (temperatur managemen
(GWL) e, relative t decisions
for humidity,
Tropical wind speed,
Peatland rainfall, and
Forest previous
Using ground
Machine water level)
Learning neural
network, the
ground
water level
is predicted
by machine
learning.

2|Page
 met, such as a predefined depth or the inability to
further split the data.
III.Methodology
INPUT AND OUTPUT SPECIFICATIONS
Required Algorithms
Input Requirements
The algorithms used in the provided code are:
The Random Forest algorithm, like many machine
1. Random Forest Classifier
learning algorithms, requires specific input data to
2. Decision Tree Classifier train and make predictions. The key input
requirements for the Random Forest algorithm
3. Support Vector Machine (SVM) include:

4. K-Nearest Neighbours (KNN) Classifier Feature Matrix:

The primary input is a feature matrix that
5. Logistic Regression Classifier
represents the training data. Each row of the
Explanation of Algorithms matrix corresponds to an instance or observation,
and each column corresponds to a feature or
1. Random Forest Classifier attribute. The matrix should be numeric and can
include both continuous and categorical features.
Random Forest is a powerful and versatile
machine learning algorithm that has gained Target Variable (Response Variable):
widespread popularity for its effectiveness in both
classification and regression tasks. It belongs to For supervised learning tasks (such as
the ensemble learning category, which involves classification or regression), there must be a target
combining the predictions of multiple models to variable or response variable. This variable
enhance overall performance. Introduced by Leo represents the outcome or label that the algorithm
Bierman in 2001, the Random Forest algorithm is aims to predict. The target variable should be
known for its ability to handle complex datasets associated with each instance in the training data.
and deliver robust results. Classification proposed Training Data:
by Eugene Kleinberg.
Random Forest requires a sufficiently large and
Firstly, the algorithm starts with the process of representative training dataset. The dataset should
bootstrapped sampling. Multiple random subsets, cover a diverse range of scenarios and patterns
called bootstrap samples, are generated by that the algorithm can learn from. A larger dataset
randomly selecting instances with replacement is generally beneficial, but the quality of the data
from the original training dataset. This step (e.g., absence of outliers or missing values) is
introduces diversity among the subsets, ensuring equally important.
that each decision tree in the ensemble is trained
on a slightly different variation of the data. Categorical Encoding (if applicable):

Next, the construction of decision trees takes If the dataset includes categorical features, they
place. For each bootstrap sample, a decision tree is need to be appropriately encoded. Random Forest
grown by recursively partitioning the data based can handle categorical variables, but they often
on randomly selected features at each node. The need to be converted into numerical
randomness in feature selection prevents representations. Common methods include one-
individual trees from becoming overly specialized hot encoding or label encoding.
to certain features, reducing the risk of overfitting. Validation and Test Sets:
The trees are grown until a specified criterion is
It is crucial to split the dataset into training,
validation, and test sets. The training set is used to

3|Page
train the Random Forest model, the validation set Trained Random Forest Model:
is employed for hyperparameter tuning, and the
The most fundamental output is the trained
test set is used to evaluate the final model's
Random Forest model itself. This consists of an
performance on unseen data.
ensemble of decision trees, each capturing
Hyperparameters (optional): different aspects of the relationships within the
training data. The model retains the knowledge
While not strictly part of the input data, the
gained during the training phase, encapsulating
algorithm's hyperparameters (e.g., the number of
the patterns and decision rules learned from the
trees, maximum tree depth, etc.) are essential
input features and target variable.
parameters that can be specified by the user.
Proper selection and tuning of hyperparameters Prediction Results:
can significantly impact the performance of the
When the trained Random Forest model is applied
Random Forest model.
to new data, it produces predictions. For
Imputation of Missing Values (if classification tasks, the output typically includes
applicable): predicted class labels for each instance. For
regression tasks, the output consists of predicted
If the dataset contains missing values, it's essential
numerical values. These predictions are generated
to handle them appropriately. Random Forest can
by aggregating the individual predictions from
tolerate missing values, but imputation methods
each tree in the ensemble.
such as mean imputation or more sophisticated
techniques may be applied to address this issue. Applications:
Balanced Classes (for Classification): Logistic regression is used in various fields,
including machine learning, most medical
In classification tasks, it is desirable to have a
fields, and social sciences. For example, the
balanced distribution of classes in the training
Trauma and Injury Severity Score (TRISS),
data. Extreme class imbalance might require
which is widely used to predict mortality in
additional techniques, such as class weighting or
injured patients, was originally developed by
resampling, to ensure that the model is not biased
Boyd et al. using logistic regression. Many
toward the majority class.
other medical scales used to assess severity of
Ensuring that the input data meets these a patient have been developed using logistic
requirements is crucial for the successful training regression Logistic regression may be used to
and deployment of a Random Forest model. predict the risk of developing a given disease
Proper preprocessing, handling of missing values, (e.g. diabetes; coronary heart disease), based
and careful consideration of the characteristics of on observed characteristics of the patient (age,
the dataset contribute to the algorithm's sex, body mass index, results of various blood
effectiveness and generalization to tests, etc.) Another example might be to
predict whether a Nepalese voter will vote
Output Format Nepali Congress or Communist Party of Nepal
The output of the Random Forest algorithm is or Any Other Party, based on age, income, sex,
multifaceted and includes various components that race, state of residence, votes in previous
provide insights into the model's performance and elections, etc. The technique can also be used
predictions. The primary outputs are associated in engineering, especially for predicting the
with both the training phase, where the model probability of failure of a given process,
learns from the data, and the prediction phase, system or product. It is also used in marketing
where it makes predictions on new or unseen applications such as prediction of a customer's
instances. propensity to purchase a product or halt a
subscription, etc in economics, it can be used

4|Page
 to predict the likelihood of a person ending up probability of someone adopting a certain
in the labour force, and a business application behaviour or attitude. In psychology, logistic
would be to predict the likelihood of a regression can be applied to model the
homeowner defaulting on a mortgage. probability of an individual belonging to a
Conditional random fields, an extension of specific psychological profile based on various
logistic regression to sequential data, are used personality traits or environmental factors.
in natural language processing.
In the realm of Information Technology and
Cybersecurity, logistic regression is
employed for intrusion detection systems. By
Logistic regression finds extensive
analysing network traffic patterns, user
applications across various domains due to its
behaviours, and system logs, logistic
versatility, simplicity, and interpretability. One
regression models can identify abnormal
prominent area of application is in Healthcare
activities and predict the likelihood of a
and Medicine. Logistic regression is
security breach. This aids in early detection
frequently employed to predict the likelihood
and prevention of cyber threats. Moreover,
of a patient having a particular medical
logistic regression is utilized in sentiment
condition based on various risk factors. For
analysis of user reviews, helping businesses
instance, it can be used to predict the
understand customer opinions about products
probability of a patient developing a specific
or services by predicting whether a review is
disease, such as diabetes or heart disease,
positive or negative.
considering factors like age, BMI, and family
medical history. In epidemiology, logistic Logistic regression is also extensively used in
regression is utilized to analyse and model the Environmental Science and Ecology. For
risk factors associated with the occurrence of example, it can be applied to model the
diseases within populations. probability of species occurrence or habitat
suitability based on environmental variables
In Marketing and Business, logistic
like temperature, precipitation, and land cover.
regression plays a crucial role in customer
Conservationists use logistic regression to
churn prediction. Companies leverage this
assess the impact of various factors on
algorithm to assess the probability of a
endangered species' survival probabilities,
customer discontinuing their services based on
aiding in the development of effective
factors like usage patterns, customer feedback,
conservation strategies.
and billing history. Additionally, logistic
regression is employed in credit scoring In Finance and Economics, logistic
models to evaluate the risk of default for loan regression is applied to predict events such as
applicants. By considering variables such as bankruptcy, stock price movements, or the
credit score, income, and debt-to-income ratio, success of a financial product. For credit risk
financial institutions can make informed assessment, logistic regression models are
decisions about whether to approve or deny a used to evaluate the probability of a borrower
loan application. defaulting on a loan. In econometrics, logistic
regression helps analyse the impact of
Within the Social Sciences, logistic regression
independent variables on the probability of an
is widely used for behavioural analysis and
economic event, such as the likelihood of a
predicting outcomes. Sociologists may use
company going public.
logistic regression to understand factors
influencing voting behaviour, predicting the In Education and Educational Research,
likelihood of an individual participating in a logistic regression is employed to study factors
particular social activity, or even exploring the affecting educational outcomes. Researchers

5|Page
 may use logistic regression to predict the decision trees, neural networks, support
probability of student success or dropout based vector machines, or ensemble methods are
on variables like socioeconomic status, trained on the pre-processed data to learn
previous academic performance, and the underlying patterns and relationships
attendance. This aids in identifying between the input features and
interventions to support students at risk. groundwater levels. The dataset is
In conclusion, logistic regression's broad typically split into training and testing sets
applicability makes it a valuable tool across to evaluate the performance of the trained
diverse fields, contributing to informed models.
decision-making and predictive modelling in
areas ranging from healthcare and finance to 5. Model Evaluation: The performance of
social sciences and environmental studies. the trained models is evaluated using
appropriate evaluation metrics such as root
Working mean square error (RMSE), mean absolute
Groundwater level prediction using machine error (MAE), coefficient of determination
learning typically involves the following steps: (R-squared), or others, depending on the
specific requirements of the application.
1. Data Collection: Relevant data related to
Cross-validation techniques may also be
groundwater levels, as well as various
employed to ensure the robustness and
factors affecting them such as
generalization capability of the models.
precipitation, temperature, land use, soil
properties, groundwater abstraction, etc., 6. Model Selection and Optimization:
are collected from different sources. This Different machine learning algorithms and
data may be historical records obtained model configurations are compared based
from monitoring stations, remote sensing on their performance metrics, and the best-
data, or other relevant sources. performing model is selected for further
optimization if necessary. Hyperparameter
2. Data Preprocessing: The collected data
tuning techniques such as grid search or
undergoes preprocessing to handle missing
random search may be used to fine-tune
values, outliers, and inconsistencies. This
the model parameters for improved
step may involve techniques such as
performance.
imputation, outlier detection,
normalization, and feature scaling to Trained models are rigorously evaluated
ensure the quality and consistency of the using appropriate performance metrics,
data. such as root mean square error (RMSE),
mean absolute error (MAE), coefficient of
3. Feature Selection and Engineering: determination (R-squared), and Nash-
Features that have a significant impact on Sutcliffe efficiency (NSE). Cross-
groundwater levels are identified and validation techniques, such as k-fold cross-
selected for model training. Additionally, validation or time-series splitting, ensure
new features may be created through the generalizability of the models across
feature engineering to capture complex different temporal and spatial contexts.
relationships or temporal dependencies in Model validation involves comparing
the data. predicted groundwater levels against
observed measurements to assess the
4. Model Training: Machine learning model's accuracy and reliability.
algorithms such as regression models,

6|Page
 7. Prediction and Deployment: Once the Overall, the process of groundwater level
model is trained and evaluated prediction using machine learning involves
satisfactorily, it can be used to make collecting, preprocessing, and analysing data to
predictions on new or unseen data. The build predictive models that can forecast future
trained model can be deployed as part of groundwater levels based on various influencing
an operational system for real-time factors. These models can help stakeholders make
monitoring and prediction of groundwater informed decisions and implement proactive
measures for sustainable management of
levels, providing valuable insights for
groundwater resources.
decision-making in water resource
management, environmental monitoring,
and other related domains.
Groundwater level prediction, a crucial aspect of
8. Real-time Prediction and Monitoring: hydrogeological analysis, encompasses a
Deployed models are utilized for real-time multifaceted process driven by interdisciplinary
approaches. Initially, comprehensive data
prediction of groundwater levels, enabling
collection is imperative, involving factors such as
stakeholders to anticipate changes and
precipitation patterns, temperature fluctuations,
fluctuations in water resources. Integration
land use characteristics, soil properties, and
of remote sensing data, IoT sensors, and historical groundwater levels. These variables
advanced monitoring technologies serve as input features for predictive models,
provides additional spatial and temporal facilitating the understanding of complex
context to model predictions, enhancing hydrological dynamics.
their accuracy and utility. Continuous
monitoring and validation of model Machine learning algorithms play a pivotal role in
groundwater level prediction by leveraging
predictions against ground truth data
historical data to discern patterns, relationships,
enable adaptive management strategies in
and trends. Techniques such as Random Forest,
response to changing environmental
Decision Trees, Support Vector Machines (SVM),
conditions and anthropogenic influences. K-Nearest Neighbours (KNN), and Logistic
9. Interpretation and Decision Support: Regression are commonly employed due to their
efficacy in handling diverse datasets and capturing
Transparent and interpretable model
nonlinear dependencies.
outputs are essential for effective decision
support in groundwater management. Preprocessing steps precede model training,
Visualization tools, dashboards, and encompassing data cleaning, normalization, and
interactive interfaces facilitate feature engineering to enhance model
stakeholders' understanding of model performance and interpretability. This phase
predictions and their implications for water involves addressing missing values, handling
resource allocation, land use planning, and categorical variables through encoding schemes
like one-hot encoding, and scaling numerical
environmental conservation efforts.
attributes to ensure uniformity and facilitate model
Additionally, uncertainty quantification
convergence.
techniques help quantify and communicate
the inherent uncertainties associated with Once trained, predictive models undergo rigorous
groundwater level predictions, enabling evaluation using metrics such as accuracy,
stakeholders to make informed decisions precision, recall, and F1-score to assess their
under uncertainty. efficacy in capturing groundwater level
fluctuations. Cross-validation techniques mitigate

7|Page
overfitting concerns and ensure robustness across and support vector machines, excel at capturing
diverse datasets. nonlinear relationships and temporal dependencies
inherent in groundwater data. Through iterative
Real-time prediction involves deploying trained
training on historical observations, these models
models on new data, enabling timely insights into
learn to discern subtle correlations between input
groundwater level dynamics. Integration of remote
variables and groundwater levels, enabling them
sensing data and advanced sensors further
to make informed predictions about future trends.
enhances prediction accuracy by providing spatial
and temporal context to model predictions. Prior to model training, extensive data
preprocessing is undertaken to ensure the quality
Interpretability and transparency are paramount,
and integrity of the input data. This involves
allowing stakeholders to comprehend model
handling missing values, scaling numerical
outputs and make informed decisions regarding
features, encoding categorical variables, and
groundwater resource management. Furthermore,
possibly feature engineering to extract meaningful
continuous model refinement and adaptation are
insights from the raw data. Furthermore, feature
essential to account for evolving environmental
selection techniques may be employed to identify
conditions and anthropogenic influences on
the most influential variables driving groundwater
groundwater systems.
fluctuations, thereby improving model efficiency
In essence, groundwater level prediction is a and interpretability.
collaborative endeavour intertwining hydro
Once trained and validated, predictive models are
geological knowledge with data-driven insights
deployed to make real-time predictions of
facilitated by machine learning techniques.
groundwater levels based on current
Through iterative refinement and interdisciplinary
environmental conditions and forecasted inputs.
collaboration, accurate and reliable predictions
Integration with remote sensing data from
empower stakeholders to make informed
satellites or ground-based sensors can provide
decisions, ensuring sustainable management of
additional spatial and temporal context, enhancing
vital groundwater resources.
the accuracy and resolution of predictions.
Groundwater level prediction operates at the Continuous monitoring and feedback mechanisms
intersection of hydrogeology, data science, and allow for model refinement and adaptation over
environmental monitoring, drawing upon a diverse time, ensuring that predictions remain accurate in
array of methodologies to elucidate the complex the face of changing environmental conditions and
dynamics of subsurface water systems. At its core, human interventions.
the process begins with the collection of
Interdisciplinary collaboration is key to the
comprehensive datasets encompassing
success of groundwater level prediction efforts,
meteorological parameters, geological
with hydrogeologists, data scientists,
characteristics, land use patterns, and historical
environmental engineers, and policymakers
groundwater observations. These datasets serve as
working together to interpret model outputs and
the foundation for developing predictive models
translate them into actionable insights for
aimed at forecasting future groundwater levels
groundwater management and conservation. By
with accuracy and reliability.
harnessing the power of advanced computational
Machine learning algorithms serve as the techniques and domain expertise, groundwater
backbone of groundwater level prediction, level prediction endeavours contribute to the
harnessing the power of computational techniques sustainable stewardship of precious water
to analyse vast datasets and uncover intricate resources for future generations.
patterns hidden within. Techniques such as
Data flow diagram & flow chart
ensemble methods like Random Forest and
Gradient Boosting, along with neural networks

8|Page
Dataset

Date Precipitation Tempe Relative Wind Solar Elevation Soil Land Use Ground
(mm) rature Humidity Speed Radiation (m) Type water
(°C) (%) (m/s) (W/m^2) Level (m)
01-01-2022 5 10 70 3 150 100 Loamy Forest 10.2
02-01-2022 8 9 75 4 130 110 Sandy Grassland 10
03-01-2022 3 12 68 5 160 105 Clayey Agriculture 9.8
04-01-2022 0 15 60 6 180 115 Loamy Grassland 9.5
05-01-2022 2 14 65 4 170 105 Sandy Forest 9.3

9|Page
 issues such as biased predictions or
unreliable model performance.
 Complexity of hydrological processes:
Advantages
Groundwater systems are inherently
 Improved Accuracy: Machine learning complex, influenced by various
algorithms can analyse large and complex hydrological, geological, and climatic
datasets, capturing nonlinear relationships factors. Machine learning models may
and patterns that traditional statistical struggle to capture the intricate
methods may overlook. This leads to more relationships and dynamics within these
accurate predictions of groundwater levels, systems, leading to oversimplified or
enabling better-informed decision-making inaccurate predictions.
in water resource management.
 Limited interpretability: Many machine
 Early Warning Systems: By continuously learning algorithms, especially complex
monitoring environmental variables and ones like deep learning models, lack
groundwater levels, machine learning interpretability. This means that while they
models can detect anomalies and forecast may provide accurate predictions, it can be
potential changes in groundwater levels. challenging to understand the underlying
This early warning capability helps factors driving those predictions, hindering
mitigate risks associated with groundwater the ability to gain insights into
depletion, contamination, or flooding, groundwater dynamics and management
allowing for timely interventions and strategies.
adaptation strategies.
 Model overfitting and generalization:
 Adaptability to Dynamic Environments: Groundwater systems often exhibit
Groundwater systems are influenced by nonlinear behaviour and complex temporal
various natural and anthropogenic factors and spatial patterns. Machine learning
that can change over time. Machine models trained on historical data may
learning models can adapt to these overfit to noise or specific patterns in the
dynamic environments by continuously training data, limiting their ability to
learning from new data, updating their generalize to new or unseen conditions,
predictions, and capturing evolving especially in the presence of changing
patterns and trends in groundwater environmental conditions or anthropogenic
dynamics. influences.

 Integration with Sensor Networks:

Advances in sensor technologies allow for
the collection of real-time data on
groundwater levels, precipitation, soil
moisture, and other relevant variables.
Disadvantages
 Data availability and quality: One
significant challenge is the availability and
quality of data required for training
machine learning models. Groundwater
level data may be sparse, unevenly
distributed, or prone to errors, leading to

10 | P a g e
 stakeholders to anticipate changes, assess risks,
and make informed decisions regarding water
allocation, land use planning, and environmental
Result: conservation efforts. Moreover, real-time
monitoring and adaptive management strategies
facilitated by predictive modeling enhance
resilience to changing environmental conditions
and anthropogenic influences.
However, it's essential to acknowledge the
inherent uncertainties associated with groundwater
level prediction, stemming from limitations in
data quality, model assumptions, and
environmental variability. Addressing these
uncertainties requires ongoing research efforts
aimed at improving data collection methods,
refining modeling techniques, and enhancing the
interpretability and transparency of model outputs.
In summary, while groundwater level prediction
offers valuable insights into the behavior of
subsurface water resources, its successful
application relies on a holistic approach that
integrates hydrogeological expertise with
advanced data analytics techniques. Through
collaborative research, monitoring, and decision
support initiatives, stakeholders can work towards
sustainable management and conservation of
groundwater resources for the benefit of present
and future generations.
V.References
IV.Conclusion  Q. W. Khan et al., "Predictive Modeling of
In conclusion, groundwater level prediction Water Table Depth, Drilling Duration, and
represents a critical aspect of hydro geological Soil Layer Classification Using Adaptive
research and water resource management, serving Ensemble Learning for Cost-Effective
as a valuable tool for understanding, monitoring, Percussion Water Borehole Drilling,"
and sustainably managing groundwater resources. in IEEE Access, vol. 11, pp. 76703-76721,
Through the application of advanced data 2023.
analytics techniques, including machine learning  L. Li et al., "Estimation of Ground Water
algorithms, predictive models can effectively Level (GWL) for Tropical Peatland Forest
capture the complex dynamics of groundwater Using Machine Learning," in IEEE Access,
systems, facilitating timely and accurate vol. 10, pp. 126180-126187, 2022.
predictions of water levels.
 S. Choudhary and M. Pundir, "Prediction
By leveraging diverse datasets encompassing of Ground Water Level in Prosopis
meteorological, hydrological, and environmental Juliflora Ecosystem using Machine
variables, predictive models can provide valuable Learning," 2023 Second International
insights into groundwater behavior, enabling

11 | P a g e
 Conference on Electronics and Renewable
Systems (ICEARS), Tuticorin, India, 2023,
pp. 1362-1366.
 S. Behera, D. Menon, G. V. Shenoy and J.
M. Suresh, "Suggestion of Appropriate
Crops Based on Rainfall and Underground
Water Analysis," 2023 3rd International
Conference on Intelligent Technologies
(CONIT), Hubli, India, 2023, pp. 1-6.
 C. G. Raju, V. Amudha and S. G,
"Comparison of Linear Regression and
Logistic Regression Algorithms for
Ground Water Level Detection with
Improved Accuracy," 2023 Eighth
International Conference on Science
Technology Engineering and Mathematics
(ICONSTEM), Chennai, India, 2023, pp. 1-
6.
 B. Mishra, K. Chauhan and R. Prasad,
"Machine learning opportunities in water
scarcity problem," International
Conference on Green Energy, Computing
and Intelligent Technology (GEn-CITy
2023), Hybrid Conference, Iskandar
Puteri, Malaysia, 2023, pp. 445-449, doi:
10.1049/icp.2023.1817.

12 | P a g e