Application of k-NN and Naive Bayesian classifier in Machine Learning

Application of k-NN and Naive Bayesian classifier in Machine Learning

• 1.
  APPLICATION OF K-NNAND NAÏVE BAYESIAN CLASSIFIERS IN MACHINE LEARNING Submitted by: Eshan Juneja 9911103457
• 2.
  Machine Learning  Subfieldof computer science  Evolved from study of pattern recognition and AI  Algorithms that can learn from and make predictions about data  Examples: Search engines, character recognition
• 3.
  Types of problems Supervised  Unsupervised  Reinforcement
• 4.
  k-NN classifier  Basicclassifier  Easy to grasp and implement  Working:  Given a set of data with labels.  For every new element, compare with existing labels.  Compare with “k” most similar pieces of existing elements  Assign class
• 5.
  Naïve Bayesian classifier Based on probability theory  Is called “naïve” because of naïve assumptions  Working:  We have probability equation of piece of data belonging to class 1 and that of class 2  For new element, if probability equation for class 1 is higher, it belongs to class 1. Else class 2.
• 6.
  Findings based onthe data set used  No. of images in training set: 1934  No. of images in test set: 947  No. of features in each image 1024 (32X32)
• 7.
  Findings based onthe data set used  For Naïve Bayesian:  Run time = O(10^6)  Error rate = 0.074974 with 71 errors in 947 cases  For k-NN:  Run time = O(10^9)  Error rate = 0.021119 with 20 errors in 947 cases
• 8.
  Conclusion  Naïve Bayesianis quicker but less accurate  k-NN is more accurate but takes more time  Both algorithms work well within acceptable error rates
• 9.
  Future work  Openended project, has lot of future scope:  More optimized algorithms  Algorithms for not just binary images, but complex images as well  Adding and comparing more modern algorithms  A fully automated text recognition system  Integration with neural networks  100% accuracy and no errors
• 10.
  Thank you!