02-02-2011, 09:15 PM
A COMPARISON OF ARTIFICIAL NEURAL NETWORKS AND CLUSTER ANALYSIS FOR TYPING BIOMETRICS AUTHENTICATION
The typing biometrics can be used as a further transparent layer which increases the security of the user authentication. The timeperiod between the individual keystrokes is used as the measure of the typing pattern of the individual which in turn is used to authenticate the user. a fully trained Multi Layer Perceptron (MLP) can be used to produce the weights and thus study the individuals typing pattern. It can also be seen as a cluster of measurements which is used to differentiate that from that of the other users.
Cluster Analysis
In cluster analysis, the pattern vectors in some sense belonging together because of similar characteristics is grouped together. With this technique, relatively homogeneous groups can be formed. The members are similar to one another within the group. But they are highly externally heterogeneous and hence not similar to the members of the other cluster.
The modified K-means cluster algorithm is a modified form of the the partitional cluster algorithm . In this technique, to reduce the clustering criterion we start with a random initial pattern and then iteratively assign these patterns to the clusters.
Modified K-means Cluster Algorithm
The problem of clustering is formally defined as:
"Given p patterns in an n dimensional metric space, determine a partition of patterns into K groups, or clusters".
http://www.kyb.mpg.de/publications/pdfs/pdf3425.pdf
The typing biometrics can be used as a further transparent layer which increases the security of the user authentication. The timeperiod between the individual keystrokes is used as the measure of the typing pattern of the individual which in turn is used to authenticate the user. a fully trained Multi Layer Perceptron (MLP) can be used to produce the weights and thus study the individuals typing pattern. It can also be seen as a cluster of measurements which is used to differentiate that from that of the other users.
Cluster Analysis
In cluster analysis, the pattern vectors in some sense belonging together because of similar characteristics is grouped together. With this technique, relatively homogeneous groups can be formed. The members are similar to one another within the group. But they are highly externally heterogeneous and hence not similar to the members of the other cluster.
The modified K-means cluster algorithm is a modified form of the the partitional cluster algorithm . In this technique, to reduce the clustering criterion we start with a random initial pattern and then iteratively assign these patterns to the clusters.
Modified K-means Cluster Algorithm
The problem of clustering is formally defined as:
"Given p patterns in an n dimensional metric space, determine a partition of patterns into K groups, or clusters".
http://www.kyb.mpg.de/publications/pdfs/pdf3425.pdf