Self-Adaptive Attribute Value Weighting for Averaged One-Dependence Estimators

Limin Wang,Minghui Sun,Yang Liu,Jie Chen

doi:10.1109/access.2020.2971706

Abstract

Of numerous proposals for weakening the attribute independence assumption of Naive Bayes, averaged one-dependence estimators (AODE) learns by extrapolation from marginal to full-multivariate probability distributions, and has demonstrated reasonable improvement in terms of classification performance. However, all the one-dependence estimators in AODE are assigned with the same weight, and their probability estimates are combined linearly. This work presents an efficient and effective attribute value weighting approach that assigns discriminative weights to different super-parent one-dependence estimators for different instances by identifying the differences among these one-dependence estimators in terms of log likelihood. The proposed approach is validated on widely used benchmark datasets from UCI machine learning repository. Experimental results show that the proposed approach achieves bias-variance trade-off and is a competitive alternative to state-of-the-art Bayesian and non-Bayesian learners (e.g., tree augmented Naive Bayes and logistic regression).

Highlights

Bayesian network (BN) [1]–[4] provides a powerful tool for knowledge representation and inference under conditions of uncertainty
The addition of augmented edges to the topology of Naive Bayes (NB) resulted in Bayesian network classifier (BNC), such as tree-augmented NB (TAN) [9] and k-dependence Bayesian classifier (KDB) [10], that achieved significant advantage over NB in terms of classification performance while retaining the simplicity and efficiency
The datasets can be divided into two groups, small datasets having less than 2k instances and relatively large datasets having more than 2k instances, and up to one million instances

Summary

INTRODUCTION

Bayesian network (BN) [1]–[4] provides a powerful tool for knowledge representation and inference under conditions of uncertainty. Jiang et al [6] proposed to discriminatively assign each attribute a specific weight for each class. Yu et al [8] assumed that highly predictive attribute values should be strongly associated with the class but not correlated with other attribute values, and different weights were assigned to attribute values by computing the difference between relevance and average redundancy. We argue that overfitting to the testing instance will help improve rather than harm the generalization performance, and the significance of each SPODE should vary while classifying different instances, especially for highly predictive SPODEs. Yu et al [4] considered the specific characteristics of each testing instance and adjusted the weights to different SPODEs adaptively by computing the correlation between the root attribute value and the class.

BACKGROUND

TARGETED AODE

Transform testing instance x to pseudo training dataset

EXPERIMENTS AND RESULTS

CONCLUSION