مقایسه روش نظارتی catboost با نیمه نظارتی از روش pu learning با الگوریتم label spreadingو رسم نمودارهای مرتبط

---

شرح پروژه

PU learning, short for Positive-Unlabeled learning, is a type of machine learning paradigm used in situations where labeled data is scarce or expensive to obtain. It falls under the category of semi-supervised learning.

In typical supervised learning, models are trained on labeled datasets where each data point is assigned a label indicating its class or category. However, in many real-world scenarios, obtaining labeled data can be costly or time-consuming. PU learning addresses this issue by working with datasets that have only positive (P) and unlabeled (U) examples, lacking explicitly labeled negative (N) instances.

The main objective of PU learning is to leverage the available positive samples and a set of unlabeled samples to build a classifier that can accurately distinguish between positive and negative instances. This is done by treating the unlabeled data as a mixture of both positive and negative samples and training the model to differentiate between the known positive instances and the unknown combination of positives and negatives.

There are different approaches to PU learning, some of which include:

1. Two-step approach: This involves first estimating the probability of each unlabeled sample being a positive instance and then training a classifier based on these estimated probabilities.

2. PU Bagging: This method involves creating multiple training sets by bootstrapping positive samples and randomly selecting unlabeled samples. Models are trained on each set and then combined, similar to ensemble learning techniques.

3. PU Learning with Confidence: This approach involves assigning confidence scores to the unlabeled data and using these scores to create a more reliable training set.

PU learning techniques aim to mitigate the lack of negative labeled data by effectively leveraging the available positive instances and utilizing the unlabeled data to improve the model's performance in distinguishing between positive and negative classes.

It finds applications in various fields where labeled data is scarce, such as in bioinformatics, fraud detection, information retrieval, and more.

مهارت ها و تخصص های مورد نیاز

---

---