A Feature-Optimized Machine Learning Approach for Early Prediction of Brain-Eating Amoeba Disease

Abstract

Brain-Eating Amoeba Disease, medically known as Primary Amoebic Meningoencephalitis (PAM), is a rare but extremely fatal infection caused by the amoeba Naegleria fowleri. The disease progresses rapidly, and its early symptoms closely resemble common illnesses such as fever, headache, nausea, and vomiting, which often results in delayed diagnosis and treatment. Consequently, the survival rate of affected patients remains very low. This paper presents a feature-optimized machine learning approach for the early prediction of Brain-Eating Amoeba Disease using patient clinical symptoms and exposure history. Due to the unavailability of real-world public datasets, a synthetic dataset derived from medical literature is used for experimental analysis. Feature optimization techniques including Chi-Square testing, correlation analysis, and Recursive Feature Elimination (RFE) are applied to select the most relevant attributes, improving prediction accuracy while reducing computational complexity. Several machine learning algorithms — Logistic Regression, Naive Bayes, Decision Tree, Support Vector Machine, and Random Forest — are implemented and evaluated. The proposed model (Optimized RF + Ensemble) achieves the highest accuracy of 91.5% with an ROC-AUC of 0.95, demonstrating the effectiveness of feature optimization in enhancing predictive performance for early detection of PAM.

Introduction

This paper presents a power quality solution for electric vehicle (EV) charging stations using a Shunt Active Power Filter (SAPF) controlled by a Synchronous Reference Frame (SRF) method.

The motivation is that large-scale EV charging introduces significant harmonic distortion into the electrical grid due to diode-bridge rectifiers used in chargers. While individual chargers may meet standards like IEEE 519-2014 (THD < 5%), multiple chargers operating together can collectively exceed acceptable harmonic limits, leading to issues such as poor power factor, transformer heating, and grid instability.

To address this, the study designs and simulates an SRF-based SAPF in MATLAB/Simulink for a multi-charger EV station. The system includes three EV charging units, with one dynamically switching on and off to simulate real-world fluctuating load conditions. The SAPF uses a voltage source inverter (VSI), DC-link capacitor, and filter inductors to inject compensating currents at the point of common coupling (PCC).

The control strategy is based on the SRF method, where load currents are transformed into a rotating d–q reference frame using Park’s transformation. In this frame, the fundamental component becomes a steady DC value, allowing harmonic components to be isolated using low-pass filtering. These harmonics are then converted back into three-phase compensating currents, which the SAPF injects to cancel distortion in real time.

The literature review shows that while active power filters and SRF control are well established, there is limited work specifically addressing multi-unit EV charging stations with dynamic load variation, which this study targets.

Conclusion

This paper presented a feature-optimized machine learning system for the early prediction of Brain-Eating Amoeba Disease (Primary Amoebic Meningoencephalitis). The system was implemented using Python in the Google Colaboratory environment and applied to a synthetic clinical dataset derived from published medical literature. Five machine learning algorithms were implemented and evaluated on the same dataset. Feature optimization techniques including chi-square testing, correlation analysis, and Recursive Feature Elimination reduced the input dimensionality from 13 to 9 features while improving model performance. Experimental results confirm that the Proposed Model (Optimized RF + Ensemble) achieves the best performance, with 91.5% accuracy, an F1-score of 90.5%, and an ROC-AUC of 0.95. The most important predictive features identified are freshwater exposure history, stiff neck, confusion, and elevated CSF pressure — consistent with clinical guidelines for PAM. The proposed system demonstrates that machine learning can be a valuable tool for early warning in rare disease detection, even in the absence of large real-world datasets.

References

[1] P. Rajpurkar et al., \"Deep Learning for Rare Neurological Disease Diagnosis,\" Nature Medicine, vol. 28, no. 3, pp. 145–158, 2022. [2] Z. Obermeyer and E. J. Emanuel, \"Predicting the Future: Big Data, Machine Learning, and Clinical Medicine,\" New England Journal of Medicine, vol. 375, no. 13, pp. 1216–1219, 2016. [3] T. Chen et al., \"Random Forest Classification for Rare Tropical Infectious Diseases,\" IEEE Journal of Biomedical and Health Informatics, vol. 25, no. 7, pp. 2401–2410, 2021. [4] X. Liu et al., \"Feature Selection Strategies for Clinical Predictive Models: A Systematic Review,\" Journal of the American Medical Informatics Association, vol. 30, no. 4, pp. 671–688, 2023. [5] A. Sharma and R. Gupta, \"Synthetic Medical Dataset Generation for Rare Disease Machine Learning,\" PLOS ONE, vol. 19, no. 2, e0298421, 2024. [6] J. Park et al., \"Ensemble and Kernel Methods for Neurological Infection Classification,\" Computers in Biology and Medicine, vol. 165, 107345, 2023. [7] Centers for Disease Control and Prevention (CDC), \"Naegleria fowleri — Primary Amebic Meningoencephalitis (PAM),\" [Online]. Available: https://www.cdc.gov/parasites/naegleria/ [Accessed April 2026]. [8] L. Breiman, \"Random Forests,\" Machine Learning, vol. 45, pp. 5–32, 2001.

Copyright

Copyright © 2026 Ms. Dora Satya Saraswathi, Ms. Nulu Anitha Rajeswari, Ms. Vura Lakshmi Devakshi Sindhu, Ms. Pabbineedi Amrutheswari. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.