Page 125 - Ai_C10_Flipbook
P. 125

The model forms two main clusters:
                    • Cluster 1: Regular transactions that follow typical patterns.
                    • Cluster 2: Suspicious transactions that deviate significantly from normal behaviour.
                 Based on these clusters, the bank can flag transactions in Cluster 2 for further investigation, without needing prior
                 labels for fraudulent activity.
                                                                                     Regular Transactions
                                                                                     Suspicious Transactions
                               Total No.   Total No. of                          80
                  Transaction
                               of Regular   Suspicious
                      ID                                                         60
                              Transactions  Transactions     Unsupervised
                                                               Learning         Transaction Frequency  40
                                                                Model            20


                                                                                 0
                                                                                       50      100     150    200     250
                                                                                              Transaction Amount ($)
                 This example illustrates how unsupervised learning helps uncover hidden patterns in unlabelled data, enabling
                 more effective decision-making.
                              Task                                                        21 st  Century   #Technology Literacy
                                                                                                  #Media Literacy
                                                                                              Skills

                   Identify the model: Supervised or Unsupervised?
                   Case 1: Email Classification
                   A system is trained to classify emails as either "Spam" or "Not Spam" based on labelled email data.
                   Case 2: Movie Recommendation System
                   An OTT streaming platform, groups users based on their viewing patterns and recommends movies without
                   predefined categories.
                   Case 3: Customer Segmentation
                   A retail company uses customer purchase histories to group customers into clusters based on buying behaviour,
                   with no predefined labels.
                   Case 4: Social media with tagged pictures
                   Social media platforms recognise your friend in a photo by analysing an album of tagged pictures.


                 Reinforcement Learning

                 Reinforcement Learning is a type of machine learning where a model learns through trial and error to make the
                 best decisions in a given environment. It interacts with its surroundings, receiving rewards for correct actions and
                 penalties for mistakes, gradually improving over time. The goal is to maximise cumulative rewards by learning
                 from experience and adapting to new situations.
                 For example, a robot learning to walk starts with random movements. When it takes a correct step, it gets a reward;
                 if it falls, it receives a penalty. Over time, the model learns which movements keep it balanced and eventually walks
                 efficiently.
                 Let us take an example of Reinforcement Learning:
                 You show the machine an image of a ball and ask it to predict
                 the object. Initially, the machine predicts it to be a globe. Since
                                                                                           Reinforcement Model     Globe
                 this is  incorrect,  it receives negative feedback.  The machine
                 then adjusts its understanding and learns that the object is not
                 a globe. When the same image is provided again, the machine
                 predicts it to be an orange. This is also incorrect, so it receives negative feedback again.
                                                                          Advanced Concepts of Modeling in AI   123
   120   121   122   123   124   125   126   127   128   129   130