Skip to main content

Table 7 Results copied from Guo and Berkhahn [3], \(\text {MAPE}\) is mean absolute percentage error, \(\text {EE}\) is entity embedding \(\text {KNN}\) is K-Nearest Neighbors, in all cases using entity embedding for encoding gives lower error, transfer learning employed for all models except neural network where direct learning is employed

From: Survey on categorical data for neural networks

Method

\(\text {MAPE}\)

\(\text {MAPE}\) (with \(\text {EE}\))

KNN

0.290

0.116

Random forest

0.158

0.108

Gradient boosted trees

0.152

0.115

Neural network

0.101

0.093