On Performance Evaluation of Mining Algorithm for Multiple-Level Association Rules based on Scale-up Characteristics

Various methods for mining association rules atmultiple conceptual levels focusing on different sets of dataand applying different thresholds at different levels havebeen proposed in literature. These are ML_T2L1,ML_T1LA, ML_TML1, and ML_T2LA. It has beenobserved that these algorithms show higher processing timeand processing cost as well as need large amount of memoryspace. This paper focuses on the comparative performanceevaluation of the ML_TMLA algorithm that generatesmultiple transaction tables for all levels in one database scanwith that of ML_T2L1 and ML_T1LA algorithms. Theperformance study has been conducted on different kinds ofdata distributions (three synthetic and one real dataset) andthresholds, which identify the conditions for algorithmselection. The Tool used for the experimental andcomparative evaluation of the proposed algorithm withother algorithms is the AR Tool. It has been concluded thatthe ML_TMLA algorithm performs better than all thealgorithms mentioned above.