Research on the Combined Prediction Model of Residential Building Energy Consumption Based on Random Forest and BP Neural Network

Table 1

Comparison of the advantages and disadvantages of machine learning algorithms.

Algorithm	Advantages	Disadvantages
DT [17]	Simple structure; suitable for handling large amount of data; fast running speed	Not easy to deal with missing data and prone to overfitting; ignore the association between attributes in the data set
KNN [17]	No requirement for data distribution, faster training phase	Not easy to find the relationship between features; large calculation amount and slow speed
SVM [18]	Solve small sample and nonlinear problems; better handling of high-dimensional data; better generalization ability	Poor interpretation of the high-dimensional mapping ability of kernel functions, especially radial basis kernel functions; more sensitive to missing data values; longer training time
BPNN [17, 18]	Strong learning ability; strong robust and fault-tolerant to noisy data; can handle nonlinear problems well	Difficult to determine the network structure; more parameters; objectiveness of the selection of training data
RF [19]	Can handle higher dimensional problems with higher prediction accuracy; insensitive to noisy data and less prone to overfitting	Belong to the black box model; difficult to explain the internal operation mechanism