This work introduces a data classification algorithm based on multi-length attractors of decimal First Degree Cellular Automata (FDCAs). The proposed model uses the basins of multi-length attractors FDCAs to form distinct attractor configuration spaces that correspond to different data classes. Since only the information of the labeled attractors are enough to classify any new data instance, this method drastically improves the time and space requirements of existing cycle-based classification approaches whereas keeping the benefits of the traditional attractor based approaches. Some selection criteria are identified to select a set of non-chaotic candidate CAs which always generate only multi-length attractors. Experimental results demonstrate that multi-length attractors FDCAs as models achieve excellent classification accuracy with execution time comparable to conventional machine learning methods, thereby validating their potential as efficient data classifiers.