Biosensor Based Detection of Early Spread of Vector Borne Disease with Personalized Treatment Strategies Using Machine Learning

Human beings are considered the pinnacle of God's creation, constantly striving to improve their lifestyles through advancements in science and technology, which have made life easier. Despite these advancements, challenges persist in protecting human life from evolving diseases transmitted through humans, animals, and environmental factors. Medical science has developed diagnostic systems to identify and treat diseases, playing a crucial role in saving lives. With the advent of AI and ML, intelligent machines now monitor health parameters and patient status continuously. The field of vector-borne disease research, however, remains in its infancy, requiring extensive studies on vectors, environmental factors, and the development of early detection and prevention technologies to prevent outbreaks [1, 2].

Vector-borne diseases, transmitted through bites of infected vectors like mosquitoes, ticks, and fleas, pose a significant threat to global public health. The complex interplay of environmental factors, vector dynamics, and human behavior makes early detection and effective treatment challenging. As these diseases spread, innovative approaches are needed to identify outbreaks early and tailor treatment strategies. This research harnesses machine learning to detect early disease growth and design personalized treatments. Traditional surveillance methods struggle with these diseases' dynamic nature, while ML can analyze large, diverse datasets, improving early intervention and patient outcomes. The focus is on developing robust predictive models and tailoring treatment plans based on individual profiles [3, 4]. By integrating complex data sources and using advanced ML algorithms, we aim to create a framework for early outbreak detection and personalized treatment recommendations for vector-borne diseases. This research could revolutionize treatment approaches, enhancing therapeutic outcomes and minimizing adverse effects. Real-time data from IoT devices and wearable technologies will enable continuous monitoring, providing a proactive, dynamic approach to public health management.

This research aims to advance precision medicine for vector-borne diseases and strengthen healthcare resilience. Our research presents opportunities for developing ahead-of-time public health strategies based on machine learning methods to defend vulnerable groups against growing vector-borne diseases [5]. Human hosts and vectors together with pathogens create complex disease transmissions that result in major health system strains across the globe. Malaria together with dengue and Zika and Lyme borreliosis diseases cover wide geographic areas while displaying varying transmission patterns. Practices of traditional disease detection together with treatment methods create delays that cause higher illness severity leading to more deaths.  Machine learning (ML) technologies achieved new levels of data analysis speed and accuracy through their advent. The research goals utilize machine learning technology to achieve vector-borne disease early diagnosis and customized therapeutic approaches which combat persistent public health crises [6].

Our goal is to merge different datasets that contain environmental elements together with patient health information in order to develop a complete predictive model for disease outbreaks and intervention customization. The immediate need to conduct this research emerges because vector-borne diseases intensify due to climate change alongside urbanization and increasing international mobility. The current disease control techniques prove inadequate so healthcare needs to transition toward personalized disease prevention approaches. Our first objective involves building predictive models through ML algorithms to analyze patterns among historical data [7].

The purpose of these models is to detect early warning signs of disease outbreak by studying vector populations with environmental factors together with human behavior elements which allows public health interventions to occur promptly and with precise targeting. The method of adopting a proactive approach proves essential to stop transmission patterns and limit major disease outbreaks. Treatment strategies should be individualized through research that considers genetics as well as lifestyle and medical background components for every patient. Through this therapy method clinical outcomes achieve peak performance while reducing adverse reactions to improve patient health. To enhance real-time access, we integrate data from IoT devices and wearable technologies for continuous monitoring and early outbreak identification, facilitating timely interventions.

1.jpg

Figure 1. Artificial intelligence and robotics

This research harnesses machine learning to revolutionize the detection and treatment of vector-borne diseases [8, 9]. By combining predictive modeling with personalized treatment strategies, we aim to develop a resilient healthcare system to address evolving challenges in the 21st century [10].

1.1 Application of machine learning in healthcare

Healthcare has become a major application area for machine learning, driven by advances in technology, new medical devices, and drugs. This sector not only saves lives but also provides employment to millions globally. Information technology plays a significant role, with smart, internet-connected medical devices monitoring multiple patients and alerting medical teams in real-time. Machine learning is increasingly accepted in healthcare, becoming a vital tool for healthcare staff and doctors [11, 12]. It analyzes vast amounts of data on diseases, offering valuable suggestions for treatment represented in Figure 2.

2.jpg

Figure 2. The illustration presents various aspects: All reported cases from 2004 to 2015 are thoroughly displayed under four categories which include annual breakdown alongside state and county data and the types of constructed facilities with reported infections. 

Blue color marks the time period between 2004 and 2005 and the green color indicates the years from 2014 to 2015. The illustration gives special attention to infection cases that occurred during 2004. The depiction focuses mostly on 2005 events to show how disease trends spread from south to north through highlighted areas with distinct gauge markings.

Table 1. Comparison of related work

After exploring several established techniques for classifying vector-borne diseases, identified significant gaps in innovation. To address these challenges, I have proposed a novel method titled "Detection of Early Spread of Vector Borne Disease with Personalized Treatment Strategies Using Machine Learning." This approach aims to revolutionize disease classification by leveraging advanced machine learning techniques for early detection and tailored treatment strategies.Top of FormBottom of Form

3.1 Prediction model for diagnosis

The proposed prediction model contains distinct modules that include Data Transformation followed by Data Preprocessing and then Dataset Partitioning and finally Model Development completed with Prediction. The Data Transformation module exists to retrieve stored data then restructure raw or semi-structured data into relational databases. The data warehouse provided CSV files or database query access which the author used for data extraction. The second module emphasizes record refinement while data normalization takes place based on the prediction model requirements. The preprocessing initiative stretches across all dengue fever characteristics by conducting proper scaling while also purifying data and removing outlier points. The resulting prediction model dataset achieves high quality through a well-defined systematic process which removes all errors. The third module of partitioning divides data between training and testing data by maintaining a 70:30 ratio. The prediction model develops based on training data until its evaluation phase uses test data. In the fourth module of the workflow teachers train the predictive model through an operation that utilizes both training and testing databases.

The team implements Gaussian Naïve Bayes Classifier within the prediction model after finishing these steps. External source information is processed by the Prediction Module through its designed system to create predictions. ThePrediction Module allows users to conduct early detection in addition to type-classification of dengue fever. This module serves two purposes including early detection of dengue through analyzing symptoms. The predictive system uses clinical reports for classification between minor and severe dengue cases. Patient records are collected by the prediction module through mobile applications and web applications as well as application software. There is a preprocessing step that follows patient data reception before the data moves to data cleaning. The system conducts patient record evaluations through either symptom-based analysis or symptom-based and clinical report analysis. The module forecasts dengue at its initial stages when the system receives only symptom data. Medical cases receive a confirmed dengue diagnosis status when symptom and clinical report data are combined in the input data. The module establishes both the normal or severe classification of dengue infection for every case. A previous explanation details the function of all modules incorporated in the vector-borne disease model.

3.2 Prediction model for vector-borne disease

The following section explains the complete operation of the prediction diagnostic framework while explaining its functions specifically for detecting vector-borne diseases starting with Dengue. The system incorporates five main functional blocks that start with data conversion then progress to data preparation before dividing the dataset between training and test sets for constructing a model and the final step is prediction. The system contains two fundamental data sets including historical data collection and newly obtained data. A local server functions as the repository for historical data storage together with its management system. The investigator selected this particular local server as their data platform. The storage system saves historical data in MySQL databases along with .csv file type. The information derives from predictions about dengue cases as well as records of rule derivations and raw data obtained from clinical laboratories. The information comes from mobile applications and web applications for users and from clinical laboratories. The next procedure requires extraction of features from existing records together with newly obtained data. The analysis utilizes descriptive statistics to make a connection assessment between different features. The imported dataset gets split into training and test sections on proportions of 70 and 30 percent respectively. The model passes an evaluation test during its development stage before undergoing result assessment. When development is complete a new data processing system with prediction capabilities becomes ready for deployment.

3.3 Data transformation module

This portion identifies the procedure for transforming raw data along with unstructured data into an analysis-ready format. Figure 5 shows how the eight-step process of data transformation takes place according to the procedure. Python pandas stores semi-structured data received from different storage sources in a new dataset.

5.jpg

Figure 5. Data transformation

Data cleanup demands specific methods to eliminate useless columns that exist in the dataset during its second procedural step. The third step of data cleaning processes removes string formatting along with alphanumeric characters from the dataset. String data gets converted into readable text which gets saved into proper columns of the dataset during this phase. The cleaned data is prepared for additional analysis after finishing its purification process.

During step 4 and for analysis purposes the transformation process moves forward by converting all categorical data into numerical values. The conversion takes place through scales that specify the particular function requirements. The prediction model contains specific encoding measurement scales to convert categorical data attributes including fever levels with values of high, medium, low and zero.

The timestamp data undergoes correct format conversion in the fifth step. The modified data completes storage in a dataset for preprocessing and normalization while Figure 5 illustrates this process (step 6).

Multiple essential actions must be taken to convert raw data records into a semi-structured vector-borne disease dataset. Regular expressions $\mathrm{E}\left(D_{\text {raw }}\right)$ work to parse and arrange raw data $D_{\text {raw }}$ in order to extract diagnostic criteria and symptoms as well as geographic information. The extracted data passes through normalization $\mathrm{N}\left(\mathrm{E}\left(D_{\text {raw }}\right)\right)$ which standardizes information while proper formatting dates and unit measurements must be unified when possible. The normalized data moves through the structuring process $\mathrm{S}\left(\mathrm{N}\left(\mathrm{E}\left(D_{\text {raw }}\right)\right)\right)$ by becoming arranged into standardized formats which often use JSON as an example. A formatted approach allows for systematic data analysis and makes it possible to use machine learning algorithms for vector-borne disease forecasting and management techniques. The transformation process comprises essential phases which establish a foundation for improving data usage effectiveness and quality as a groundwork for healthcare decisions and preventive treatment.

3.4 Data preprocessing, function scaling and normalization module and feature selection

The predictive model depends on this section where two critical processes are conducted: feature engineering and data preprocessing. Primary features from the dataset are extracted in the feature engineering module's first phase which enables establishing inference rules for the planned predictive model as Figure 6 illustrates. At the start of the process the transformation module executes the loading of the dataset. Significant features related to early detection and classification of dengue are recognized during this stage. The researcher establishes diagnostic rules to identify dengue disease in the next stage. After an analysis of selected features' relationships the prediction model retains those features which demonstrate dependency. During this step the elements are resized before data values get converted to the necessary format and data type.

6.jpg

Figure 6. Data preprocessing and feature scaling

3.5 Data pre-processing

At the beginning of the process the transformation module operates to fetch the dataset. The database provides data retrieval after selecting vital features in the second stage of data processing and cleaning. The researcher proceeds to examine and address absent data values and replace them with correct elements in the third processing step. Identifying and eliminating duplicate records stands as the fourth step while the fifth step requires outlier detection to remove them. The sixth step of normalization operates on every dataset component to maintain uniformity. All modifications within this section get safely stored on the server at the conclusion of the seventh step. Normalization and Min-Max Scaling and standardization are part of the preprocessing stage.

3.5.1 Min-max scaling

The Min-Max technique applied to functions transforms variable values into a standardized 0 to 1 range by using minimum and maximum values found in the input data. All feature values get proportionally modified through this process to match the range from 0 to 1. The following formula represents the Min-Max Scaling formula:

$\bar{x}=\frac{x-\min (x)}{\max (x)-\min (x)}$

3.5.2  Standardization

Standardization transforms data features through an adjustment process which sets their central value (or mean value) at zero and applies a standard deviation value of one. The normalization method known as Z-score produces standardized values which are written as "z" in the output. The standardized method will maintain an equivalent data distribution skew that exists in the original data set.

$\frac{z=x-\mu}{\sigma}$

3.5.3 Normalization

Normalization expands all vectors in the coordinate space while shrinking or lengthening their lengths. Every feature selected during feature selection and correlation matrix analysis has been chosen by the researcher in this study. The Dataset needs this process for proper creation. Scaling features accurately ensures that the database formation remains precise. Features that undergo scaling help predict models avoid errors while improving their final performance standards. The next process demands the researcher to analyze both the scale and normalized values of features while focusing specifically on binary classification in the research. The investigator adjusts the scales of patient features particularly fever variables to use zero as the minimum value and one as the maximum.

3.6 Divide dataset module

The preprocessed dataset was divided into three distinct sets to ensure a comprehensive evaluation of the model's performance. Specifically, 70% of the data was allocated for training purposes, allowing the model to learn from a substantial portion of the dataset. An additional 15% of the data was designated for testing, providing an initial measure of the model's accuracy and effectiveness [19]. The validation segment occupied 15% of the data to conduct final assessments which refined the model following its application to unforeseen data. The organized methodology provides reliable data assessments throughout all stages of testing [20].

The Gaussian Naïve Bayes Classification machine learning algorithm powers the model construction module which enables the completion of the envisioned diagram shown in Figure 7.

7.jpg

Figure 7. Normalization

3.7 Model building module

The Naive Bayes model uses probabilistic foundations together with an assumption of independent features to identify vector-borne diseases. Initial disease probability discovery occurs by analyzing historical datasets that contain symptoms together with region data alongside patient demographics. Training enables the model to determine both feature likelihood when assigned to each class along with initial probabilities of class occurrence. New data input leads the model to apply Bayes' theorem for calculating the posterior class probabilities based on the observed features. The model performs its classification by estimating the predicted disease from the most probable class. Despite its simplifying assumption of feature independence, Naive Bayes can effectively categorize vector-borne diseases by utilizing these probabilistic calculations to make informed classifications based on the data's patterns and distributions [21].

$P\left(y \mid x_1, x_2, \ldots, x_n\right)=\frac{P(y) \prod_{i=1}^n P\left(x_i \mid y\right)}{P\left(x_1, x_2, \ldots \ldots, x_n\right)}$

where, $P\left(y \mid x_1, x_2, \ldots, x_n\right)$ is the posterior probability of class y given features $x_1, x_2, \ldots, x_n, \mathrm{P}(\mathrm{y})$ is the prior probability of class $\mathrm{y}, P\left(x_i \mid y\right)$ is the likelihood of feature $x_i$ given class y. $P\left(x_1, x_2, \ldots, \ldots, x_n\right)$ is the evidence, which acts as a normalizing constant.

3.7.1 Feature selection

Feature selection is the process of identifying and selecting the most relevant features (variables) from a dataset to improve model performance by reducing overfitting and enhancing interpretability.

Information Gain (IG): Measures the reduction in entropy or uncertainty about the class label C when a feature $x_i$ is known:

$I G\left(x_i, C\right)=H(C)-H\left(C \mid x_i\right)$

where, $\mathrm{H}(\mathrm{C})$ is the entropy of the class labels and $H\left(C \mid x_i\right)$ is the conditional entropy of C given $\left(x_i\right)$.

Chi-Squared Test: Assesses the independence between each feature ( $x_i$ and the class C . The chi-squared statistic is calculated as:

$x^2\left(x_i, C\right)=\sum_{x_i \in { values }\left(\left(x_i\right)\right.} \frac{\left(o_{x_i}-E_{x_i}\right)^2}{E_{x_i}}$

where, $O_{x_i}$ is the observed frequency of class C for feature $x_i$, and $E_{x_i}$ is the expected frequency under independence.

Recursive Feature Elimination: Recursive Feature Elimination (RFE) can be applied to Naive Bayes by recursively training the classifier and removing the least important features based on model coefficients or feature ranking scores

1. Initialization: Start with all p features in the dataset $\mathrm{X}=\left\{X_1, X_2, \ldots X_p\right\}$.

2. Train Machine Learning Model on the current set of features X. Obtain feature important scores from model.

3. Rank the features based on their importance scores.

4. Eliminate the least important feature(s) from X based on the ranking. Update X to exclude the eliminated features.

5. Repeat steps 2-4 iteratively until the desired number of features is selected or a stopping criterion (e.g., a predefined number of features, no improvement in model performance) is met

Regularized Naive Bayes:

These techniques typically involve adding regularization terms to the Naive Bayes model to penalize certain feature coefficients, thereby influencing feature selection

Objective Function: The objective function typically includes a regularization term $\lambda$ added to the log-likelihood or error function, aiming to penalize complex models or features that may overfit:

Objective $=$ Log-Likelihood $+\boldsymbol{\lambda}$. Penality

Common penalty terms include L1L1L1 (Lasso) or L2L2L2 (Ridge) norms applied to feature coefficients or priors:

L1 Regularization (Lasso):

$\lambda \sum_{j=1}^p\left|\beta_j\right|$

L2 Regularization (Ridge):

$\lambda \sum_{j=1}^p \mid \beta_j^2$

where $\beta_j$ are again the coefficients of features $X_j$

During model training, the regularization term $\lambda$ is tuned to balance between fitting the data well and keeping the model simple.

Impact: Regularization encourages sparse solutions by shrinking less important feature coefficients towards zero, effectively performing feature selection.

The selection criteria for features relies on the coefficients which remain nonzero after regularization to identify important factors for model operation.

3.8 Prediction module

The chapter describes the methods for vector-borne dengue forecasting through the developed prediction model. The prediction model provides diagnosis for two different types of dengue patients who either do or do not report their symptoms. The model for prediction will serve healthcare providers by evaluating data that patients or laboratory technicians feed into its system. The first phase of data collection and examination allows patients who do not present NS1 or IgM reports to share their records for dengue fever diagnosis. Subsequent subsections detail the techniques and methods for diagnosing dengue fever in this context. Healthcare professionals including patients and medical staff present records with NS1 test results and blood antibody levels (IgM, IgG) together with platelet levels to patients.

The prediction system efficiently identifies confirmed dengue fever types and stages through a process covering both initial diagnostic and classification procedures.

Classification: Given new data instances $x_{n e w}$, the classifier predicts the most probable class $\hat{y}$ using Bayes' theorem:

$\hat{y}=arg~max P(y) \prod_{i=1}^n P\left(x_i \mid y\right)$

The model determines class probabilities then predicts the class having the maximum probability as $\hat{y}$. Theresearcher presents a complete description of their work to develop a machine learning-driven vector-borne dengue disease prediction system which focuses on Indian stage classification and early diagnosis methods. The model comprises five modules: Data Transformation, Data Preprocessing, Feature Scaling and Normalization, Dataset Partitioning, Model Building, and Prediction. The first module executes multiple data transformation steps which start with raw data loading followed by redundant column removal and string formatting elimination and alphanumeric character clearing alongside categorical data conversion to numbers and timestamp adjustment.

The second module encompasses data preprocessing, feature scaling, and dataset normalization. Feature scaling comprises the content of module three while the fourth module describes preparing a proposed Gaussian Naïve Bayes classification model. Dengue-infected patient diagnostic tool uses a modified classification approach to speed up dengue detection and diagnostic stage identification. The researcher built four algorithms which identify dengue exposure during its early stages with the categories including Dengue-DF for normal dengue and the subsequent classifications of Dengue-DHF and Dengue-SD for severe stages. The current stage of development enables the proposed vector-borne disease prediction model to diagnose dengue infections at different stages prior to taking it through tests and validation steps using real-time data.

The subsequent chapter delves into the testing and validation of this proposed model for vector-borne dengue disease.

8.jpg

Figure 8. Correlation matrix of inference rule

The inference rule contains seven vital components needed to identify dengue early (Figure 8) that follow the 2009 WHO guidelines. The 2009 WHO guidelines state that two simultaneous symptoms among fever, nausea, rash, abdominal pain, vomiting and retroorbital pain should trigger dengue infection assessment. Each dengue symptom points toward an average of six other symptoms according to the correlation matrix evaluation. This data displays co-relationships between rash related to nausea together with vomiting being connected to pain thus demonstrating a strong probability for such connections. Patients with fever along with pain tend to develop nausea and rash and vomiting according to the correlation matrix findings.

Table 2. Measuring table

In machine learning, a property refers to a variable or characteristic of an object. Every element receives a numerical value that spans from maximum to minimum possible limits. The Element Scale describes data intake for specific elements through the established maximum and minimum numerical values. The model responds according to the input values assigned to its elements within this framework. The procedure of defining the boundaries for an element falls under normalization. Between standardization and normalization and minimum-maximum scaling machine learning uses these practices most frequently as feature scaling approaches. The table provides additional information about these methods as shown in Table 2.

A person's excessive body temperature serves as the key indicator of dengue disease while other symptoms like headache and nausea and vomiting and retroorbital pain and joint pain appear during the infection period. The researcher established a classification system for fever which divides the temperature condition into four categories: Normal, Low, Medium, and High for prediction development purposes. The index value gets established from data points connected to fever input variables described in Table 2.

Performance Metrics

Performance metrics are essential in evaluating and refining machine learning models. They provide quantitative measures to assess model effectiveness, highlight weaknesses, and guide improvements

Accuracy: The ration of correctly predicted instances to the total instances.

$\mathrm{A}=\frac{T P+T N}{T P+T N+F P+F N}$

Precision (P): Ratio of correctly predicted positive observations to the total predicted positives.

$\mathrm{P}=\frac{T P}{T P+F P}$

Recall (R): Ratio of correctly predicted positive observations to all the observations in the actual class.

$\mathrm{R}=\frac{T P}{T P+F N}$

F1-Score: The weighted Averages of Precision and Recall.

$F_s=2 * \frac{{ Precision } * { Recall }}{ { Precision }+ { Recall }}$

These metrics facilitate clear communication of model performance to stakeholders, ensuring that the chosen model aligns with the project's goals and constraints.