Predicting Loan Repayment

Classification and Hypothesis Development

• Classification aims to predict whether we will be paid back based on various attributes (a, c, i, e, o, u)
• Multiple hypotheses can be developed to predict payment, including:
  • Income is high
  • Income is high and no Criminal record
  • Complex combinations of attributes (e.g., Address is known, NOT Old, OR Unemployed)

Nearest Neighbor Approach

• Nearest neighbor approach predicts payment based on the most similar example in the training data
• Distance is calculated between the new instance and each training instance
• The nearest neighbor's payment status is used for prediction

k-Nearest Neighbors Approach

• k-Nearest Neighbors (k-NN) approach predicts payment based on the k most similar examples in the training data
• A vote is taken among the k neighbors to determine the payment status

Perceptrons Approach

• Perceptrons approach uses weights to indicate the importance of each attribute in predicting payment
• Scores are calculated by summing the weighted attributes, and a threshold is set to determine payment status

Reinforcement Learning

• Reinforcement learning involves exploration vs. exploitation tradeoff
• Exploration involves trying new options, while exploitation involves sticking with known good options
• Reinforcement learning is often studied in Markov Decision Processes (MDPs)

Bayesian Approach to Learning

• Bayesian approach assumes a prior distribution over the long-term behavior of a route (e.g., A)
• The prior distribution is updated based on new observations, allowing for learning and adaptation

Learning in Game Theory

• Learning in game theory involves multiple agents learning simultaneously
• The environment is changing from one agent's perspective, making it challenging to make decisions
• Taking the average of past observations may not be the best approach in game theory