Reciprocating compressors are important machinery used in the oil and gas industry but are prone to wear and failure due to their many moving parts. Traditional maintenance methods like reactive and scheduled inspections often miss early signs of faults, which can lead to unplanned downtime, costly repairs, and safety risks. Predictive maintenance offers a more proactive approach by identifying potential faults before failure can happen. This study explores a data-driven approach for fault detection using snapshot data from an open-source dataset of a two-cylinder, single-stage reciprocating air compressor. The dataset contains sensor data and health status labels for four components: bearings, radiator, water pump, and discharge valve. A multi-layer perceptron (MLP) model was built to classify the condition of each component as either healthy or unhealthy. The architecture consisted of 20 input neurons, two hidden layers, and 4 neurons in the output layer which corresponds to each component. The model was tested using accuracy, loss, precision, and recall metric for evaluation. Bayesian optimization was applied to reduce the size of the parameters for a more efficient performance. The final model achieved an average macro F1-Score of 0.991 and a 1.0 recall for all the components. These results show that snapshot-based fault classification has potential to assist operators with early fault detections and better maintenance decisions.