Recursive Shrinking Toward Effective Cluster Isolation for Robust Electronic Noses

In electronic noses (e-Noses), the employed sensors’ responses consist of overlapping clusters leading to inaccurate analysis. Larger intra-cluster distances and smaller inter-cluster distances within the dataset cause overlapping clusters. The lack of well-separated clusters hinders pattern recognition techniques from excelling and requires effective isolation for optimal performance. This work proposes recursive shrinking towards effective cluster isolation utilizing the synergy of principal component analysis and the bisection method. The clusters shrink towards their centers on each recursion by optimizing an objective function, effective inter-cluster distance (EICD). Overlapping characterizes negative EICD. The experimental findings demonstrate the effectiveness of the suggested approach on a dataset that includes responses from five different alcohol categories: 1-octanol, 1-propanol, 2-butanol, 2-propanol, and 1-isobutanol. The used dataset exhibits highly overlapped clusters with negative-valued EICD. Clusters of 1st, 2nd, 3rd, and 4th alcohol overlap with subsequent peers (i.e., 1-2, 3, 4, 5; 2-3, 4, 5; 3-4, 5; 4-5) and achieve negative EICD. Recursive shrinking produces completely isolated clusters with positive EICD values. The results depict the effectiveness of isolation numerically and graphically.