| With the rapid development of large data and the Internet, the data shows the characteristics of diversity, rapid growth and so on. To make better use of these data and to discover the laws showed in the form of the model is a meaningful research direction. ID3 algorithm is a classic algorithm in machine learning. ID3 algorithm ensures that the depth of decision tree is smaller but not the number of leaves is less. However it ensures that the depth of decision tree is smaller but not the number of leaves is less. RLID3 algorithm is proposed in this thesis for that. Specifically, the algorithm idea is that using Decision Tree Optimization ratio as the index of selecting split properties. The experiment result shows that RLID3 is better than ID3.In practical applications, sometimes the data comes in batches, all data can't be got at once. In most cases,the traditional scheme is using the new data to build a new model.However, some of the old data still are valuable, this scheme is not appropriate. Incremental learning algorithm can use the new data on the basis of the old model for further learning to avoid these problems partly. Incre_RLID3 algorithm is one of Incremental learning algorithm.Incre_RLID3 algorithm is shown to have higher precision.Ensemble learning is an effective method to improve prediction precision. It is proved by a large number of experiments. PAR_WT algorithm is one of ensemble learning. It trains a plurality of the single model by using RLID3 algorithm, and then assembles model's forecasting results to increasing precision. Considering the advantages of incremental learning and ensemble learning, an incremental integration algorithm called Incre_RLID3_ENM is proposed. Incre_RLID3_ENM algorithm trains a plurality of the single model by using Incre_RLID3 algorithm, and then assembles model's forecasting results. Experiments show that the precision of Incre_RLID3_ENM algorithm is higher than generating a single model using Incre_RLID3 algorithm. |
PDF Full Text Request