Optimized Reversible Cellular Automata Based Clustering

The research optimizes reversible cellular automata based clustering technique for any high dimensional dataset. The reversible rules are characterized using the cycle structure properties of each rule to identify effective rules for clustering. This essentially reduces the rule search space for a given neighborhood size. A novel encoding technique (BiNCE Encoding) that encodes any dataset into binary form without significant data loss is also introduced for our algorithm. Finally, the algorithm and implementation is transformed into a package which is applicable on various datasets, split sizes and cluster sizes for ease of accessibility and reproducibility. While compared against the state-of-the-art methods using benchmark clustering metrics, it is shown that our algorithm is at par or beating the scores for certain datasets and settings.