This research paper focuses on the development and evaluation of an Android-based smart healthcare system

improves diagnostic accuracy for diabetes, achieving approximately 75% prediction capability. This work

contributes a viable mobile health solution that facilitates early diabetes diagnosis, improves patient

management, and enhances healthcare accessibility in similar settings, thereby promoting a paradigm shift

towards technology-driven healthcare.

The global healthcare landscape is undergoing significant transformations, driven by an escalating burden of

chronic diseases, an aging population, and persistent disparities in access to quality medical care (Zhao, Luo,

and Qiu, 2017). Traditional healthcare systems often struggle to cope with rising demand, characterized by

overburdened medical personnel, inconsistent patient record management, and limited diagnostic capabilities,

particularly in remote or under-resourced regions. This often leads to delayed diagnoses, suboptimal patient

2030 (Ali et al., 2023). In many developing countries, including Zimbabwe, the prevalence of diabetes is a

growing concern, contributing to increased morbidity and mortality. Traditional diagnostic methods often rely

on episodic clinical visits and reactive symptom management, frequently missing early indicators of the disease

or failing to provide continuous, personalized care. This reactive approach is particularly problematic in contexts

where healthcare infrastructure is limited, and access to timely interventions is scarce.

In response to these challenges, the concept of "smart healthcare" has emerged as a promising paradigm,

leveraging advancements in information and communication technologies (ICTs), mobile computing, and data

analytics to transform healthcare delivery (Sundaravadivel et al., 2018; Tian et al., 2019). Smart healthcare

systems aim to shift from a reactive, hospital-centric model to a proactive, patient-centered, and prevention-

oriented approach. Central to this transformation is the integration of data mining techniques and machine

the region, while beneficial, often focus on specific areas like maternal health (Nyati-Jokomo et al., 2020),

leaving a void in comprehensive, real-time disease diagnosis and management for prevalent chronic conditions.

Furthermore, many healthcare facilities, such as those within the Zimbabwe Defence Forces (ZDF) where this

study is primarily focused, continue to rely on paper-based record systems, impeding efficient data utilization

for proactive health interventions.

This research addresses this critical gap by focusing on the development and evaluation of an Android-based

smart healthcare system specifically designed for enhanced diabetes prediction using data mining techniques.

The primary objective is to create a mobile application that can provide preliminary diabetes diagnosis based on

patient-provided symptoms and other health metrics, offer personalized healthy diet suggestions, and facilitate

seamless interaction between patients and healthcare providers. By leveraging an ensemble machine learning

body of knowledge on the application of advanced data mining techniques in healthcare, particularly

demonstrating the efficacy of ensemble learning in predictive analytics within low-resource contexts. The

findings and the developed prototype can serve as a foundational model for broader adoption of smart health

technologies across Zimbabwe and similar developing regions, fostering a paradigm shift towards a more

intelligent, reliable, and patient-centric healthcare system.

The remainder of this paper is structured as follows: Section 2 provides a comprehensive review of related

literature on smart healthcare, data mining, and their applications in disease prediction. Section 3 details the

methodology employed, including the research design, data collection, system architecture, and machine

learning model development. Section 4 presents the key findings and analysis of the developed system's

performance, with a specific focus on diabetes prediction accuracy. Finally, Section 5 concludes the paper,

integrates advanced information and communication technologies (ICTs), the Internet of Things (IoT), mobile

computing, and data analytics to create intelligent, interconnected healthcare ecosystems (Tian et al., 2019;

Sundaravadivel et al., 2018). This evolution is driven by the imperative to address mounting global health

challenges, including the escalating burden of chronic diseases, an aging global population, and the persistent

disparities in access to quality medical services, particularly in developing regions (Preprints.org, 2025; PMC,

2025).

Key characteristics of smart healthcare systems include context awareness, responsiveness, personalization, and

intelligence, enabling dynamic information access and real-time decision-making (Sundaravadivel et al., 2018).

These systems facilitate remote patient monitoring through wearable devices and biosensors, automate data

collection and analysis in the cloud, and enable seamless communication between patients and healthcare

often referred to as Knowledge Discovery in Databases (KDD), is the process of identifying implicit, previously

unknown, and potentially useful information from large datasets (Mohapatra et al., 2018; Han, Kamber, and Pei,

2012). In healthcare, data mining plays a crucial role in transforming raw patient data into actionable knowledge,

assisting in disease diagnosis, prognosis, treatment planning, and fraud detection (Baitharu and Pani, 2016;

Hooda, 2017).

The KDD process typically involves several iterative stages: data cleaning, data integration, data selection, data

transformation, data mining, pattern evaluation, and knowledge representation (Wibamanto, Das, and Chelliah,

2020). By systematically applying these steps, healthcare organizations can uncover hidden trends and predictive

patterns that might otherwise be overlooked by human experts, thereby enabling more informed and proactive

clinical decisions (Deshpande and Thakare, 2016). The impact of data mining in the medical field is profound,

prediction, which falls under the classification task, and recent research continues to validate the effectiveness

of various ML and DL models in diagnosing diabetes and tracking its progression (Ayoade et al., 2025).

Support Vector Machines (SVM) are powerful supervised learning models used for classification and regression

analysis. Originally designed for binary classification, SVMs have proven highly effective and adaptable to

multiclass scenarios in healthcare (Bhatia, 2019). SVMs operate by constructing a hyperplane or a set of

hyperplanes in a high-dimensional space, which optimally separates data points into different classes. The

primary objective is to maximize the margin between these classes, leading to robust classification. SVMs are

advantageous due to their effectiveness in high-dimensional spaces and their memory efficiency, as they only

use a subset of training points (support vectors) for decision function construction. However, their performance

may not always hold true in complex medical contexts), Naïve Bayes often performs remarkably well,

particularly with large datasets (Patel and Patel, 2016). Its computational simplicity makes it a fast and accurate

choice for various disease prediction tasks. Researchers globally utilize Naïve Bayes for its ability to facilitate

computation processes and its effectiveness in handling large datasets efficiently (Patel and Patel, 2016).

While SVM and Naïve Bayes are central to this study, other classification algorithms frequently applied in

identify new data forms (Patel and Patel, 2016). Regression techniques, on the other hand, are used to examine

and predict relationships between variables, often employed for assessing illness progression or survivability

(Larose and Larose, 2015). Recent advancements also include the use of deep learning models like Convolutional

Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) for analyzing unstructured medical data

(Ayoade et al., 2025).

Individual data mining algorithms, while powerful, often possess inherent weaknesses that can limit their

accuracy and robustness, especially when dealing with complex and noisy healthcare data (Kirubha and Priya,

2016). To overcome these limitations, ensemble learning methods have emerged as a superior approach.

of patient data for model training and evaluation, while the qualitative component gathered insights into user

needs, challenges with existing systems, and perceptions of the proposed mobile application.

representativeness.

informants) to gather professional insights into disease diagnosis, patient management, and the potential

understand their experiences with existing healthcare services and their needs for mobile health solutions.

Secondary data played a crucial role in the development and evaluation of the disease prediction model:

However, due to confidentiality issues and the paper-based nature of existing records, comprehensive

datasets were not readily available for direct use in model training.

Science Direct, IEEE journals, and recent preprints/journals from 2020-2025) on smart healthcare, data

mining techniques, and mobile health applications was conducted to inform the conceptual framework,

identify relevant algorithms, and understand current trends and challenges.

The sampling strategy for primary data collection combined both probability and non-probability methods:

questionnaires and a portion of the interviews, ensuring that every member of the target population had

an equal chance of being selected. This contributed to the generalizability of the qualitative findings

within the ZDF healthcare system.

doctors, dieticians) and the case study area (ZDF clinics in Harare) based on their relevance and ability

Quantitative data, derived from questionnaires and the downloaded Kaggle datasets, was analyzed using:

focusing on numerical and statistical interpretations of responses. This included demographic analysis,

descriptive statistics, and potentially inferential statistics to understand relationships between variables.

involved checking data shape, exploring data distributions, and analyzing correlations between

features (e.g., using heatmaps).

the machine learning models.

To ensure the reliability and validity of the data collected:

questionnaires. A reliability coefficient of 0.70 was set as the acceptable threshold, indicating good

internal consistency.

datasets) and multiple data sources (patients, doctors, hospital staff, Kaggle) enhanced the validity

of the findings.

use an advanced module for disease prediction (potentially integrating lab results), add new

diseases/symptoms, and view scheduled appointments.

administrators, resetting passwords, and assisting with patient and doctor registrations.

System diagrams, including sequence diagrams, system architecture diagrams, use case diagrams, and entity-

relationship diagrams, were developed to illustrate the proposed system's functionality and structure.

Ethical clearance was obtained from the Faculty of Engineering, Zimbabwe National Defence University.

Official letters were issued to relevant faculties and institutions to facilitate data collection. All study participants

were fully informed about the research's purpose, their right to withdraw at any time, and the confidentiality of

This chapter presents the findings derived from the primary and secondary data collected, focusing on the

demographic characteristics of the study population, the prevalence of diabetes within the Zimbabwe Defence

Forces (ZDF) healthcare system, and the performance of the developed Android-based smart healthcare system,

particularly its diabetes prediction capabilities. The analysis integrates insights from questionnaires, interviews,

and the evaluation of machine learning models to address the research objectives. Data is presented using

descriptive statistics, tables, and figures for clarity and ease of interpretation.

A total of 160 questionnaires were targeted, with 129 actual responses received (80.6% response rate) from

medical personnel and patients across ZDF hospitals and clinics in Harare. Additionally, 5 out of 6 targeted

interviews (83.4% response rate) with doctors and dieticians were successfully conducted. These high response

a mobile healthcare application.

within the healthy BMI range (18.5-24.9). However, a notable percentage of both males (15%) and

females (18%) had BMIs between 25-30, and 6% of males and 4% of females had BMIs between 35-40,

indicating a segment of the population at risk of obesity-related conditions, including diabetes. This

The research specifically investigated the prevalence of diabetes within the ZDF population. Findings indicated

a direct proportionality between age and the prevalence of diabetes, consistent with global trends (Alzboon,

2025; Lee et al., 2025). As individuals age, the risk of developing diabetes increases significantly. The data also

showed that males generally exhibited a higher risk of diabetes compared to females across most age groups.

This demographic insight underscores the importance of a predictive system that can identify at-risk individuals,

particularly older males, for early intervention and targeted health management.

The developed Android-based smart healthcare system prototype was designed with a 3-tier architecture (User

Interface, Business Logic Layer, Database) to ensure scalability, maintainability, and efficient data flow. It

comprises three interconnected modules:

in, view patient details (including BMI and treatment history), utilize an advanced module for disease

prediction (which could integrate laboratory results in a full implementation), add new diseases and

symptoms to the system's knowledge base, and manage scheduled appointments.

administrators to manage user accounts (enable/disable), assign administrative rights, reset user passwords,

and assist with patient and doctor registrations.

The system's design, as illustrated by the Use Case and Entity-Relationship Diagrams in the original dissertation,

ensures a comprehensive and integrated approach to healthcare management, with a specific focus on facilitating

early diagnosis and personalized care for conditions like diabetes.

NumPy), loading the dataset, checking its shape, exploring initial data patterns (e.g., data.head()), and

verifying the absence of null values (data.isnull().values.any()). Correlation analysis, visualized through

heatmaps, was performed to understand the relationships between features and the target variable, guiding

feature engineering.