Jul 2026· Asian Journal of Research in Computer Science· 0 citations
TL;DR
This framework illustrates, without any claim of clinical validity, how a leakage-safe ML pipeline and SHAP interpretability can be combined and rigorously self-audited; real patient-level data and external validation are required before any clinical inference is drawn.
Abstract
Malaria remains one of the most significant causes of morbidity and mortality in tropical and subtropical regions, and timely diagnosis is essential for effective case management. Microscopy is the traditional parasitological reference standard, while rapid diagnostic tests (RDTs) are widely used, field-deployable alternatives with product- and antigen-dependent sensitivity and specificity; both approaches are constrained by requirements for trained personnel, reagents, or equipment in resource-limited settings. This study develops and evaluates, on a simulated clinical dataset calibrated to published aggregate statistics, an explainable artificial intelligence pipeline for malaria diagnosis prediction from routinely collectable symptoms, vital signs, and haematological indices, using six machine-learning (ML) models: Logistic Regression (LR), Naive Bayes (NB), K-Nearest Neighbours (KNN), Random Forest (RF), Support Vector Classifier (SVC), and Decision Tree (DT). The Synthetic Minority Oversampling Technique (SMOTE) and Random Forest feature selection are embedded within a single leakage-safe pipeline that is refitted in every cross-validation fold. The reported best model is selected on the basis of the cross-validated F1-score rather than held-out test performance, and the test set is used exactly once for confirmatory reporting. Under this design, SVC with RF-selected features was selected (mean cross-validated F1 = 0.740), achieving a test-set accuracy of 0.906 [95% CI 0.852, 0.953], recall of 0.867 [0.667, 1.000], and ROC AUC of 0.959 [0.919, 0.988]. A paired bootstrap test found no statistically significant difference in AUC compared with the runner-up, Logistic Regression (AUC 0.956, p = 0.81). Permutation importance corroborated 8 of the top 10 impurity-based features. Parasite density, the quantity used to determine the parasitological diagnosis, was excluded from the predictor set as a precautionary safeguard against near-total label leakage. Calibration, subgroup recall by age and sex, and a class-weighting comparison are also reported. This framework illustrates, without any claim of clinical validity, how a leakage-safe ML pipeline and SHAP interpretability can be combined and rigorously self-audited; real patient-level data and external validation are required before any clinical inference is drawn.
Background Sub-Saharan Africa continues to shoulder the heaviest burden of malaria. The 2024 WHO malaria report highlighted that Africa contributed an alarming 94% of the global cases and 95% of the deaths. In the WHO African region, progress towards elimination and management of malaria is hindered by weak health systems, and lack of traditional diagnostic methods such as microscopy and malaria rapid diagnostic tests (mRDT). The primary aim of the study is to develop a machine learning (ML) ensemble model for malaria diagnosis using clinical and demographic data, tailored for resource-limited settings. Methods A retrospective study was conducted using 637 patient records from Gutu Mission Hospital and Gweru Provincial Hospital in Zimbabwe. Clinical symptoms (fever, chills, abdominal pain, headache and diarrhea) and demographic features (age, gender, residence and travel history) were analysed. Data preprocessing included handling class imbalance using Synthetic Minority Oversampling Technique (SMOTE) and feature selection using Recursive feature elimination (RFE). Seven individual ML models including Logistic regression (LR), Random Forest (RF), Decision Trees (DT), Gradient Boosting (GB), K-Nearest Neighbor (KNN), Naive Bayes (NB) and XGBoost were trained and evaluated on the malaria dataset. The individual models were further combined to build, train and evaluate ensemble models such as Bagging, Stacking, Soft Voting and AdaBoost. Model performance was assessed using accuracy, precision, confusion matrices, recall and F1score and AUR-ROC metrics. Results Clinical symptoms (chills: p=0.001, fever: p=0.003, diarrhoea: p=0.01, abdominal pain: p<0.001) were statistically significant predictors of malaria. Of the demographic factors, only travel history (p=0.02) showed significant association with malaria. Among the seven individual ML models, GB achieved the highest predictive performance (Accuracy = 0.94), followed by RF (Accuracy = 0.94%) and XGBoost (Accuracy = 0.93%). The stacking ensemble model outperformed all individual ML models and other ensemble models (bagging, soft voting and adaBoost) achieving accuracy = 0.96, precision = 0.95, recall = 0.98, F1 score= 0.96 and AUC-ROC = 0.98. Conclusion This study demonstrates that ML particularly ensemble models can be used to significantly improve malaria diagnosis. The integration of these models into a web-based application could provide a scalable and accessible diagnostic tool for healthcare workers in resource limited settings.
Panashe Nyengera, H. Takawira, F. Mlambo· medRxiv· 1 citation
Beyond predictive accuracy, qualitative criteria including explain-ability, transparency, and adaptability were evaluated, further highlighting the superiority of CBR 2 over conventional black-box models.
Konan N’gatta Aimé Kouassi, Koffi Kouakou Ive Arsene, Goore Bi Tra· International Journal of Adv...· 0 citations
This study demonstrates the effectiveness of machine learning techniques in malaria risk prediction using clinical information and proposes a malaria risk prediction model using machine learning techniques based on clinical information to potentially improve early detection and treatment of malaria, ultimately reducing the burden of the disease.
Prabhat Kumar, Pragati Sahu, Smaranika Priyadarshini et al.· 2 citations
Automated malaria diagnosis from stained blood-smear microscopy is dominated by deep convolutional neural networks that are accurate but computationally expensive, poorly interpretable, and rarely validated with patient-level rigor. We present EMFE (Efficient Mathematical Feature Extraction), a five-feature framework for classifying single red-blood-cell images as parasitized or uninfected using Gray World color normalization, adaptive green-channel thresholding, morphological spot detection, and classical machine learning. Using the NIH LHNCBC malaria dataset (27,558 images from 200 patients), we evaluate Random Forest, Histogram Gradient Boosting, and Support Vector Machine classifiers under patient-grouped nested cross-validation (K_outer=20, K_inner=3), ensuring that cells from each patient remain within a single fold. The optimized Random Forest achieves 94.6% pooled out-of-fold accuracy (95% CI [93.6, 95.7]), corroborated by an untouched 40-patient holdout test (94.3%) and a patient-level permutation test (p<0.001, 1,000 permutations). Ablation experiments quantify the contribution of individual features and pipeline stages. Hardware-matched comparisons with retrained DenseNet121, ResNet50, and MobileNetV2 models assess the accuracy-efficiency trade-off. Synthetic perturbations characterize three failure modes, while explainability analysis identifies spot saturation as the dominant discriminative feature. Patient-level aggregation further quantifies sensitivity-specificity trade-offs and false-positive accumulation. These results demonstrate a statistically rigorous, interpretable, and computationally lightweight alternative to deep learning, while explicitly quantifying its limitations.
Majid Kafi, Walayat Hussain, Mousumi Karmakar et al.· 0 citations
Background: Tuberculosis (TB) continues to be one of the most pressing health issues worldwide, and in 2024, 10.6 million new cases were reported, resulting in 1.3 million deaths. However, the early and accurate diagnosis is crucial to limit transmission but the existing means are expensive, time consuming, and impractical in resource-limited environments. The objective of the present study was to build a quick and low-cost diagnostic model based on the use of blood biomarkers routinely available in the laboratory. Tuberculosis (TB) continues to be one of the more deadly diseases, particularly in the developing world. This study is aimed at developing a prototype solution where primary signs, symptoms and risk factors of TB would be identified at an early stage and machine learning (ML) predictive algorithms would be applied to them.
Methods: Data for 818 confirmed TB cases and 2,618 healthy controls were analyzed in a retrospective manner. Since the dataset is imbalanced, the ROSE technique was used to balance the training set. Seven machine learning algorithms were trained, and feature selection was done using LASSO regression and forward selection to determine the most predictive variables. SHAP analysis was used to increase interpretability of the model, and the final predictive model was implemented as an interactive Shiny web application to make it easier to be used in the clinic.
Result: The Gradient Boosting Machine (GBM) model demonstrated superior performance on the test set, achieving an area under the curve (AUC) of 0.821, specificity of 85.7%, and sensitivity of 64.9%. SHAP analysis highlighted platelet-to-lymphocyte ratio (PLR), monocyte-to-lymphocyte ratio (MLR), and platelet distribution width (PDW) as the most influential predictors. Adjusting the classification threshold to 0.24 improved sensitivity to 82.6% while maintaining acceptable specificity (58.9%), underscoring the model’s potential utility as a screening tool. The accompanying Shiny application enhances accessibility and practical deployment in clinical settings.
Conclusion: This study presents a robust and interpretable GBM-based diagnostic model leveraging routine hematological parameters to provide a rapid, low-cost TB screening tool suitable for resource-constrained environments. The model’s high specificity may reduce unnecessary confirmatory testing, while its core predictors offer biological insights into TB-associated inflammatory processes. The interactive web application facilitates integration into clinical workflows, supporting early detection and improved TB management.
Yahya Khan, Lisa Sullivan, Danial Khan et al.· World Journal of Biology Pha...· 0 citations
Hantavirus infection remains a rare but potentially fatal zoonosis, and early identification of patients at highest risk of death is essential for timely triage and resource allocation. This study developed an interpretable machine learning (ML) framework for mortality risk stratification using a global hantavirus epidemiology dataset. The dataset included both Hemorrhagic Fever with Renal Syndrome (HFRS) and Hantavirus Cardiopulmonary Syndrome (HCPS) cases. The study conducted a retrospective supervised learning analysis on 10,000 patient records, including demographic, clinical, epidemiological, and treatment variables. The binary outcome was mortality. Preprocessing included identifier removal, missing-value handling, categorical and symptom encoding, feature selection, and class-imbalance correction. Logistic regression, random forest, and extreme gradient boosting (XGBoost) models were trained and compared on a held-out test set using receiver operating characteristic-area under the curve (ROC-AUC), accuracy, precision, recall, F1-score, and Brier score. Model interpretation was planned using SHapley Additive exPlanations-based feature attribution. The cohort included 8,938 recovered and 1,062 deceased cases (mortality rate: 10.62%), comprising 6,460 HFRS cases (64.6%) and 3,540 HCPS cases (35.4%). Logistic regression achieved the highest discrimination, with an ROC-AUC of 0.858 and the highest recall for mortality detection (0.741), but modest precision (0.321) and weaker calibration (Brier score 0.119). Gradient boosting showed comparable discrimination (AUC 0.857) with better precision (0.511) and calibration (Brier score 0.076). Random forest performed less well for mortality detection, with markedly low recall (0.085). The study trained and compared interpretable classifiers using 10,000 global hantavirus cases representing both HFRS (64.6%) and HCPS (35.4%) presentations, and reported pooled performance as well as syndrome-stratified results. Severity, syndrome type, viral load category, age, and geographic setting were the most informative predictors. Mortality risk in hantavirus infection can be modeled using routine clinical and epidemiological features, but clinical deployment should prioritize sensitivity, calibration, and transparency over accuracy alone. These findings support the use of interpretable ML as a practical framework for early hantavirus risk stratification and external validation in independent cohorts.
Yetunde Enigbokan, A. Abiona, M. Ajinaja· Cureus Journal of Computer S...· 0 citations