Missing Values and Optimal Selection of an Imputation Method and Classification Algorithm to Improve the Accuracy of Ubiquitous Computing Applications

Table 8

Factors influencing accuracy (RMSE) for each algorithm (standard beta coefficient): Predictive_Mean_Imputation.

Data characteristic	trees.J48	BayesNet	SMO	Regression	Logistic	IBk
N_attributes	−.076^**	−.076^**	−.178^**	−.063^**	.123^**	.016
N_cases	−.084^**	−.049^**	.012	−.017	−.034^*	−.047^**
C_imbalance	.117^**	.242^**	.263^**	.523^**	.153^**	.198^**
R_missing	.050^*	.079^**	.043	.085^**	.080^**	.068^**
SE_HS	.223^**	.279^**	.182^**	.268^**	.322^**	.242^**
SE_VS	−.008	−.013	−.006	−.013	−.015	−.009
Spread	−.328^**	−.432^**	−.262^**	−.434^**	−.465^**	−.361^**
P_missing_dum1	−.042	−.035	−.034	−.028	−.044	−.036
P_missing_dum2	.008	.012	.004	.018	.007	.011

Note  1: N_attributes: number of attributes, N_cases: number of cases, C_imbalance: degree of class imbalance, R_missing: missing data ratio, SE_HS: horizontal scatteredness, SE_VS: vertical scatteredness, spread: missing data spread, and missing patterns: univariate (P_missing_dum1 = 1, P_missing_dum2 = 0), monotone (P_missing_dum1 = 0, P_missing_dum2 = 1), and arbitrary (P_missing_dum1 = 1, P_missing_dum2 = 1)
Note  2: RMSE indicates error; therefore, lower values are better.
Note  3: ^* < 0.05, ^** < 0.01.