Due to its various uses in different areas, machine learning has become one of the most popular fields of artificial intelligence. It can be certainly said that data classification by trained machine vision algorithms is currently used as a model in most of software types. It is obvious that the variety of methods for modeling the samples leads to different percentages of accurate detection in the data of a shared dataset. The higher is this percentage, the more valid will be the proposed method. The output of classification algorithms may cause a large extent of interference in detection. This interference can be due to the noisy data or the samples that are less frequently observed in data training phase. To boost the output of this method, we use meta-heuristic algorithms that boost the improperly-classified samples by determining the error level in predetermined assumptions. One of these methods is AdaBoost that is characterized by the major deficiency of lack of noise sensitivity and overfitting. Based on the proposed method, we can decrease the noise sensitivity and improve the overfitting problem in less frequent data by using the ant colony optimization algorithm; so that the probability of improper classification of the outliers is significantly reduced.