To date, most methods for loss reserving are still used on aggregate data arranged in a triangular form such as the Chain-Ladder (CL) method and the over-dispersed Poisson (ODP) method. With the booming of machine learning methods and the significant increment of computing power, the loss of information resulting from the aggregation of the individual claims data into accident and development year buckets is no longer justifiable. Machine learning methods like Neural Networks (NN) and Random Forest (RF) are then applied and the results are compared with the traditional methods on both simulated data and real data (aggregate at company level).