2.4.1. Confidence weighted voting (CWV) strategy
More specific predictions can be given by assembling scores from
individual models. Limitations in individual models can be avoided by
integrating scores from many models, resulting in higher overall
accuracy. Models blended in this way usually perform at least as well as
the best and sometimes better individual models. Another weighted
majority voting form is Confidence
Weighted Voting (CWV). However, rather than giving high weights to the
nearest sensors, CWV gives higher weights to those sensors that are more
likely to be right (i.e., with higher confidence of correctness). A
distributed comparison of sensing results and neighbors that share
overlapping coverage can be made for the confidence value of each
sensor. For this study, a confidence-weighted voting strategy was used
to combine the scores from individual NN, SVM, and QUEST algorithms (11,
14).
\begin{equation}
\sum_{m=1}^{M}{p_{m,j}d_{m,j}=\max_{k=1}^{K}}p_{m,j}\ \left(\sum_{m=1}^{M}d_{m,k}\right)\nonumber \\
\end{equation}\(p_{m,j}\) is the posterior likelihood for a given vector of predictor
values calculated for the kth target group by the mth base model. M is
the number of base models, and K is the number of target categories
(15).