Figure 7

Model results of one year sales for fiction and nonfiction books. We compare Learning to Place against three baseline methods (Linear Regression, KNN, Neural Network). (A) Actual vs prediction scatter plots. All three baseline methods systematically underpredict at high-end to a certain extent while this underprediction is absent in Learning to Place. (B) Quantile-Quantile plot. For fiction, on the high-end sales, neural network and Learning to Place are the closest to the ground truth (45 degree line). For nonfiction, Learning to Place is the closest to the ground truth. KNN and Linear Regression at the high-end systematically have lower predictive power for both fiction and nonfiction