Lassa fever remains a significant public health concern in Nigeria, where surveillance data can support outbreak monitoring and response planning. This study compared selected machine learning models for predicting weekly confirmed Lassa fever cases and classifying threshold-defined outbreak weeks using nationally aggregated Nigeria Centre for Disease Control and Prevention (NCDC) surveillance data from 2020 to 2025. The dataset comprised 313 weekly observations, reduced to 261 observations with 33 predictors following temporal feature engineering. A chronological validation strategy used 2020–2023 for training, 2024 for validation, and 2025 for independent testing. Regression performance was assessed using mean absolute error (MAE), root mean squared error (RMSE), mean absolute percentage error (MAPE), and coefficient of determination (R²). Classification performance was evaluated using accuracy, precision, recall, F1-score, and receiver operating characteristic area under the curve (ROC-AUC). Support Vector Regression (SVR) achieved the strongest reported regression performance, with an MAE of 6.3546, RMSE of 8.7611, and R² of 0.8148. For classification, optimized Random Forest and XGBoost models each recorded 0.9038 accuracy, 0.8095 precision, 0.9444 recall, and 0.8718 F1-score. Their ROC-AUC values were 0.9673 and 0.9641, respectively. The outbreak threshold was defined as at least 18 confirmed cases, based on the training-period 75th percentile, rather than an official NCDC declaration. These findings demonstrate that model performance varied by prediction task and suggest that temporal machine learning may complement routine surveillance. However, discrepancies in reported classification outputs require reconciliation, alongside further validation, before operational deployment.