Research case study

Explainable AI for Healthcare

Classification and ensemble-learning research designed to make clinical predictions more inspectable.

Problem

Healthcare models must balance predictive utility with transparent reasoning that clinicians can interrogate.

Approach

Data: healthcare tabular and image datasets
Model: weighted ensembles, CNNs, and Vision Transformers
XAI method: Grad-CAM and explainable ensemble analysis
Validation: class-balanced evaluation and threshold optimisation

Results

TODO: Dice score TODO: clinical validation TODO: model comparison

Figures

Grad-CAM activation overlay highlighting neural network decision regions — Grad-CAM overlays on misclassified guava leaves, field-condition subset. demo

Confusion matrix layout showing class-wise error inspection and classification performance — Confusion matrix layout for class-wise error inspection. demo

Receiver operating characteristic curve comparing model discrimination capabilities — ROC curve layout for threshold and class-separation analysis. demo

Weighted voting and stacking ensemble workflow architecture — Weighted voting and stacking ensemble workflow. demo