With the growth of internet and advance in computing technologies, large volume of data is generated and processed on daily bases. The situation is very appealing for sophisticated tools and techniques to process and get the insight of such a large data ubiquitous in every domain. Classi - cation and pattern recognition is an important technique in data mining task tond unexplored knowledge within the data. Emerging Pattern (EP) based technique is used to discover an abrupt change in across the datasets. Emerging pattern based classi ers are very e cient innd- ing patterns inherently exist among the collection of large datasets in hand. Nevertheless, the discovery of emerging patterns is a challenging and non-trivial task due to the inherent complexity of datasets. More- over, in contemporary literature plethora of work is available tond a hidden emerging pattern. Among these techniques,tree-based approach is commonly utilized for the discovery of emerging pattern. However, a tree-based approach follows greedy search technique, su ers various limitations i.e. tree size grows with problem size up to the level where tree-based computation is not cost e ective. In this dissertation, a novel approach for discovering emerging patterns has been proposed. The pro- posed approach exploited Ant Colony Optimization (ACO) technique for the exploration of high quality emerging patterns in the classi cation problem. Furthermore, in contrast to the deterministic approach used in tree-based classi ers, the proposed probabilistic method provides com- petitive classi ers. The proposed approach is capable to e ciently avoid an exhaustive search of tree-based approach and obtain comparatively better accuracy to discover strong jumping emerging patterns. The pro- posed approach has been evaluated on various benchmark datasets for accuracy and robustness. Additionally, most of the classi ers are work with discrete data only and Discretization methods are used to change continuous data into discrete before input to a particular classi er. To evaluate the performance of proposed approach on di erent discretiza- tion methods several experiments are carried out to evaluate the robust- ness. Experimental results show that proposed approach provides better performance on di erent datasets in comparison with the state-of-the-art techniques.