|
چکیده
|
In a standard classification framework, a meticulously selected collection of reliable training data is employed to develop a decision rule designed to reliably classify unlabeled cases within the test set. This method has a considerable limitation, it requires a substantial quantity of labeled cases to facilitate effective learning. Due to the manual nature of labeling, the process can be highly laborious and time consuming. For this reason, semi-supervised approaches have been developed to address these difficulties. Within this educational context, there exists a limited collection of classified instances for each class, alongside a substantial collection of uncategorized instances. The goal is to utilize the unlabelled examples in order to enhance the process of learning. The existence of unreliable labeled observations, including outliers and inaccurately labeled data, can severely impair the classifier’s performance. This danger is especially significant when the training dataset is somewhat small, as it may lack adequate information to mitigate these mistakes. The work presented here introduces a robust modification to the model-based classification framework, integrating the concepts of impartial trimming and incorporating constraints on the ratio between the maximum and minimum eigenvalues of the group scatter matrices. In order to achieve these goals, in the First chapter, we describe the classification preparations, the evaluation of the results, and the types of classification methods. The Second chapter will include the model-based classification, Semi-supervised classification and the concept of attribute and class noise, and outliers and its effect on the classification. In the Third chapter, robust model-based classification for attribute and class noise is discussed in detail and we introduce a new technique called RUPCLASS along with the parameter estimation method with the EM algorithm. In the Fourth chapter, using several simulations and analysis of heart failure data in Pakistan, we put the described methods and algorithms into practical use and evaluation. Finally, in the Fifth chapter, conclusions and future work.
|