Ai-based Intelligent Multi-disease Prediction System Using Adaptive Symptom Analysis - A Comparative Study Of Supervised Learning Algorithms For Clinical Disease Classification

• Author(s):

  Ms. Indhumathi S | Livya Grace E | Sneka V | Visvanath R

• Keywords:

  ArdSymptom-Based Disease Prediction, Supervised Machine Learning, Support Vector Machine, Multi-Class Clinical Classification, Healthcare Artificial Intelligence, Adaptive Symptom Analysis, Comparative Algorithm Evaluation

• Abstract:

  The Integration Of Artificial Intelligence Into Clinical Healthcare Settings Has Gained Remarkable Momentum Over The Past Decade, Offering Unprecedented Opportunities To Augment Diagnostic Capabilities And Extend Medical Reach To Underserved Populations. Among The Many Promising Applications Of This Integration, The Automated Identification Of Diseases From Patient-reported Symptom Profiles Stands Out For Its Potential To Democratize Access To Preliminary Healthcare Screening. This Paper Presents An Empirical Comparative Investigation Of Four Supervised Machine Learning Algorithms — Random Forest, K-Nearest Neighbors, Naive Bayes, And Support Vector Machine — Evaluated On The Task Of Predicting Diseases From Binary Symptom Feature Vectors. Experiments Were Conducted On A Structured Dataset Containing 4,920 Patient Records Distributed Across 41 Disease Categories And Encoded Using 131 Symptom Attributes. A Stratified 80/20 Train-test Partition Was Employed Alongside 5-fold Cross-validation To Ensure Reliable And Generalizable Performance Estimates. Among The Four Algorithms Evaluated, The Support Vector Machine Equipped With A Radial Basis Function Kernel Consistently Outperformed Its Counterparts, Attaining A Test Accuracy Of 99.19%, A Weighted F1-score Of 99.16%, And A Cross-validation Mean Of 99.27% — Results That Clearly Exceeded The 90% Performance Target Established At The Outset Of This Research. Beyond The Algorithmic Investigation, The Paper Describes The Development Of A Fully Functional Flask-based Web Application Named MediAI, Which Embeds The Trained SVM Model Within An Adaptive Symptom Collection Interface Capable Of Delivering Real-time Differential Diagnoses, Confidence-calibrated Predictions, And Actionable Clinical Guidance. These Findings Collectively Affirm That Classical Supervised Learning, When Thoughtfully Applied To Well-structured Clinical Data, Can Serve As A Reliable Foundation For Accessible And Scalable Disease Screening Systems.

---

Other Details

• Paper id:

  IJSARTV12I4104880

• Published in:

  Volume: 12 Issue: 4 April 2026

• Publication Date:

  2026-04-05

Download Article