The Wheel Impact Load Detector (WILD) sensor measures the force exerted by the railway wheel on the rail track. If the force exerted is beyond a threshold value, the wheel needs inspection. The Problem Solving Competition 2017 organized by INFORMS Railway Application Section, asked participants to predict the force exerted by wheel, when a currently empty rail car would be loaded in the next trip. Further, convert the predicted force value into an alert, if the value is more than the threshold value. We present machine learning models to address the above problem along with results. We label the predicted output into two classes. Records with predicted force greater than or equal to the threshold value as 1 else 0. %True Positives detected were found to be very low (0–6%), which results in poor prediction of critical event. Instead of predicting critical event based on predicted force valued, we label the actual/training records, thereby converting the prediction problem into a classification problem. We apply various classifiers to classify whether the record is critical or not. We apply filtering, data pre-sampling, cross-validation and hyperparameter tuning. As the resultant labeled data is highly imbalance, we apply Synthetic Minority Oversampling Technique (SMOTE) and Cost Sensitive learning techniques to increase True Positives and reduce False Alarms. Random Forest classifier with SMOTE turned out to be the best among all classifiers with %True Positives as 85.76%, %False Alarms as 8.27%.