Malaria remains a critical public health challenge in Nigeria, where accurate diagnosis is essential for effective
disease management and resource allocation. Discordance between rapid diagnostic tests (RDTs) and microscopy poses
significant challenges for malaria surveillance programs, potentially leading to misdiagnosis and inappropriate treatment
decisions. This study aimed to develop a machine learning model for predicting diagnostic test discordance between RDT
and microscopy in malaria surveillance data from Bayelsa State, Nigeria. A dataset comprising 2,100 monthly observations
from eight Local Government Areas spanning January 2019 to December 2024 was analyzed. The methodology incorporated
Bland Altman agreement analysis, feature engineering with climate and health system variables, and gradient boosting
classification with class weight balancing to address data imbalance. Model interpretation was achieved through SHapley
Additive exPlanations (SHAP) analysis. The Bland Altman analysis revealed a mean difference of negative 2.33 percentage
points between RDT and microscopy, with limits of agreement spanning negative 19.28 to positive 14.62 percentage points.
The LightGBM classifier achieved an area under the receiver operating characteristic curve of 0.901, with precision of 0.67,
recall of 0.74, and F1 score of 0.703. SHAP analysis identified rainfall, climate index, geographic location, and humidity as
the most influential predictors of diagnostic discordance. This study contributes an interpretable machine learning
framework for identifying conditions associated with diagnostic disagreement, potentially informing quality assurance
protocols and targeted interventions in malaria endemic regions.