This study compares several popular Machine Learning (ML) models for predicting US recessions. The models tested include gradient boosting, random forest, support vector classifier, feedforward neural network, and a custom consensus model. Predictions are made for 3, 6, 9, 12, and 18-month periods. The dataset spans from 1962 to 2023 and utilizes six leading indicators for training and testing the models. Two evaluation techniques are used to evaluate the models' performance. The first method employs a ranking system based on accuracy, precision, recall, F1, and AUROC scores. The second method uses bootstrapped AUROC scores to conduct a one-sided test on each model pair across all prediction horizons. The results indicate that gradient boosting, random forest, and consensus models outperform the logistic regression model in predicting recessions. These models show better classification capabilities across all time horizons. On the other hand, the neural network and support vector classifier models do not demonstrate as strong performance compared to the logistic regression model. Similarly, the findings from the consensus model highlight the advantage of combining predictions from multiple models. This approach results in more accurate recession forecasts than relying on the outputs of a single model.