The accelerating volume and structural diversity of data generated by healthcare systems, smart medical devices,

and enterprise platforms has rendered conventional data analysis pipelines increasingly impractical for

nonexpert users. This paper presents an extensible, machine learning-integrated analytics platform designed to

enable interactive, code-free analysis of heterogeneous medical datasets through a web-based interface. The

system accepts structured and semi-structured data in CSV and Excel formats and executes an automated

pipeline encompassing data preprocessing, descriptive statistical analysis, interactive visualization, and machine

learning — including regression, clustering, and anomaly detection — without requiring users to possess

programming skills.

maintaining an architecture designed for modular extension.

Keywords: Machine Learning; Heterogeneous Data Analysis; Big Data Analytics; Web-Based Analytics;

Automated Data Processing; Predictive Analytics; Data Visualization; Healthcare Data Systems;

ScalableAnalytics Architecture; Decision Support Systems.

The proliferation of digital health technologies has positioned data as a foundational resource for both

organizational decision-making and scientific inquiry. Data streams originating from clinical sensors, electronic

health records (EHRs), e-commerce transactions, and social media platforms are accumulating at rates that

challenge the capacity of traditional analytics infrastructure. This growth has fundamentally altered how

decisions are made across disciplines, from clinical medicine to urban systems management.

including clinicians, healthcare administrators, and research scientists. This accessibility gap prevents many

potential users from leveraging available analytical capabilities.

To address this challenge, this paper introduces a Machine Learning-Based Extensible Analytics Platform for

Heterogeneous Medical Data Analysis. The system is a web-based application supporting near code-free data

analysis. Users upload CSV or Excel files, and the platform automatically performs preprocessing, statistical

summarization, visualization, and machine learning inference through an accessible graphical interface.

The technical foundation comprises a React-based frontend, a FastAPI backend, and a Python processing layer

utilizing Pandas, NumPy, Scikit-learn, and Matplotlib. The platform contributes to the broader effort to

democratize data analysis in the context of Industry 4.0. The following sections detail relevant literature, system

design, experimental results, and directions for future work.

rendering them impractical for routine clinical use [1][2].

Recent scholarship has increasingly advocated for integrated approaches that combine big data analytics with

machine learning to improve prediction accuracy and enable more responsive decision support [3][4]. Several

web-based analytics platforms have been developed with the goal of simplifying data exploration

through visualization and limited automation. However, the majority do not provide a

comprehensive end-to-end pipeline — from raw data ingestion through preprocessing, advanced modeling, and

interactive reporting — within a single unified system [5][6].

Furthermore, most current platforms exhibit limited extensibility, constraining their adaptability to

emerging data sources and evolving requirements. The absence of a scalable, user-accessible, and architecturally

flexible medical data analytics environment motivates this work. The proposed platform bridges

well-formed CSVs and Excel files to loosely structured or semi-structured records. Managing this structural

heterogeneity at scale presents a significant analytical challenge.

Although sophisticated tools for data analysis and machine learning exist, their effective utilization demands

proficiency in programming, statistical methodology, and complex software environments. This excludes

clinicians, academic researchers, and business analysts from accessing the insights these tools can generate.

The specific deficiencies motivating this work include: difficulty integrating heterogeneous medical data sources

in a single environment; heavy coding dependencies in existing systems; the absence of automated end-to-end

processing workflows; a lag in integrating modern machine learning capabilities into accessible interfaces; and

the lack of a unified, transparent, and extensible analytics workspace suitable for non-specialist users.

Proposed System

programming background. The system consolidates data processing, machine learning, and visualization within

a unified web-based environment.

analytical pipeline. Uploaded data serve as the primary input for all subsequent processing and analytics

functions.

Preprocessing Module

Upon data ingestion, the system automatically parses and structures the input. This module performs type

inference across all columns, identifies and imputes missing or null values using configurable strategies (mean,

median, or mode substitution), and removes structural inconsistencies. The output is a cleaned, type-consistent

dataset prepared for downstream analysis.

Statistical Analysis Engine

Following preprocessing, the platform performs automated descriptive statistical analysis. Core measures —

including mean, median, variance, standard deviation, and interquartile range — are computed for all numeric

Means, DBSCAN), and anomaly detection (Isolation Forest). Model outputs are surfaced through the interface

without requiring users to configure model parameters directly.

Visualization and Dashboard Module

Analytical results are communicated through an interactive visualization layer generating bar charts, line plots,

scatter plots, correlation heatmaps, and cluster visualizations. A dynamic dashboard enables users to explore

results, filter views, and examine data distributions in real time. This visual presentation is designed to maximize

interpretability for nonspecialist audiences.

System Architecture

The backend is implemented using FastAPI, which manages incoming requests, orchestrates model execution,

and returns structured responses. The frontend is constructed in React.js, providing a responsive, state-driven

The platform was evaluated across a series of benchmark medical datasets to assess predictive accuracy,

processing efficiency, and visualization responsiveness. Experiments were conducted on publicly available

clinical datasets spanning regression, classification, clustering, and anomaly detection tasks, sourced from the

UCI Machine Learning Repository and synthetic ICU telemetry streams. All evaluations used an 80/20 train-

test split.

Table II summarizes the key performance metrics observed across representative task types. Regression

performance was assessed via the coefficient of determination (R²); classification performance via weighted F1-

score; clustering quality via the Silhouette Coefficient; and anomaly detection precision via holdout ground truth

labels.