Data Classification And Protein Function Prediction Problem Research Based On Ant Colony Algorithm

With the advent of big data, data mining and data classification have increasingly become the focus of attention to researchers. Because of the complexity, redundancy, difficulty of modeling and so on, the existing algorithms are difficult to accurately classify data. Artificial intelligence heuristic algorithms have certain advantages to deal with data classification and some other optimization problems. Artificial ant colony intelligence system has been developed to be a new intelligent algorithm by mimicking the ant colony foraging of social behavior. Ant colony algorithm contains the evolutionary learning mechanisms, such as self-organization, memory function and positive feedback, which provides new solutions and ideas for complex optimization problem. Because ant colony algorithm has superior performance on complex optimization problems, it gradually becomes a research focus.This thesis studies ant colony intelligence algorithm from the application instance of data classification problems. At first, it introduces the research status of ant colony intelligence algorithm, biological mechanism and algorithm principle. Then, we analyze data classification and protein function prediction problems and put forward the improved algorithms based on ant colony algorithm. Through the simulation experiments, the improved algorithms have the obvious advantages. In this paper, the main work is listed as follows:(1) When dealing with data classification problems, this thesis proposes a novel multiple rule sets data classification algorithm, named Ant Minermbc, based on ant colony algorithm. This algorithm proposes a novel classification model, which is composed by multiple rule sets to make up for each other, and this new model can improve the classification accuracy. Besides that, a new heuristic function is designed in this thesis to effectively avoid the over-fitting problem. At last, the weighted voting mechanism is adopted to accurately classify unknown data.(2) When dealing with protein function prediction problems, this thesis presents a new hierarchical multi-label classification algorithm, named hm Ant Minerorder, based on ant colony algorithm. This algorithm puts forward an orderly roulette selection strategy, which can clearly distinguish the pros and cons of protein data attributes and help to guide the artificial ants for searching a better solution. Besides that, a new pheromone update strategy is also designed in this algorithm to strengthen its convergence and to avoid the trapping in local optimal solution when dealing with protein function prediction problems.