Define Probably Approximately Correct (PAC) learning. What is its significance in machine learning? Given a learning algorithm with an error rate of 0.2, how many samples should be drawn to achieve a confidence level of 95% in PAC learning?

OR

(b) Explain the K-Nearest Neighbors (KNN) algorithm. How does the choice of 'K' influence the model's behavior? Given a dataset with 120 samples, each having three features, implement the KNN algorithm with Manhattan distance as the similarity measure. If K-7, find the class label for a new data point with the following feature values: Feature1=6.3. Feature2=2.5, Feature3=4.8.