Prediction Soaring Price by Decision Tree Dealing with Imbalanced Data in Trading Card Game Market

The market for used trading card games (TCG) has expanded rapidly in recent years. In the TCG market, the demand for used TCG cards tends to rise in tandem with the release of new products, driving up the prices. Unlike the regularly used market, the used TCG market has some challenges with price increases. We investigated the used market for "Magic: The Gathering," one of the TCGs, to forecast price increases. However, in terms of prediction, there are very few soaring products, almost all of which are classified as "not soaring" and they are of poor precision. Preliminary experiments also revealed that the recall and F1 values were extremely low, resulting in poor prediction performance. Dealing with imbalanced data using three handling techniques, we attempted to increase the recall in this research's predictions of TCG's skyrocketing price. As a result, the handling method improved recall of the methods' ability to predict price increases. Furthermore, we attempted to compare the recall of three methods for dealing with imbalanced data (oversampling, undersampling, and cost-aware learning approach). The outcome demonstrates that the costaware learning approach and random sampling were particularly effective handling methods.