The convergence property of cellular automata makes them effective for pattern classification and elementary cellular automata have already proven their efficacy. This work uses binary cellular automata with 5-neighborhood architecture. In this work, the candidate cellular automata for classification obey the following properties: (i) cycles of length one only, (ii) stable cyclic behavior and (iii) moderate number of cycles. These three properties are investigated on balanced and unbalanced rules both. In case of unbalanced category, rules with at most 25% 0 and rules with at most 12.5% 1 are investigated for a set of CA sizes. Finally, the model is trained and tested using the rules, some of which perform at par with the existing classifiers by exhibiting optimal classification performance. This makes to the observation that the 5-neighborhood rules can be used as efficient classifiers. Experimental results establish that balanced and unbalanced both rules perform efficiently, and interestingly, rule 1114110 shows its efficacy as classifier on three different real data sets.