Artificial Intelligence

Outcomes

Insights

• Patterns found that advise business decisions

• Reveal the unknown

  • E.g.: Frequently bought together

Utility

• Can be use repetitively to solve similar tasks, but may not provide insight to solutions

• Faster and better solutions to promblems

Data

Structured Data

• If the cells are numbers/bools, they can be inputed directly to the machine learning models

• If the cells are text, they often need encoding

  • One hot encoding: Often used for categories that do not have a natural order

  • Ordinal encoding: Often for categories that have a natural order

Unstructured Data

• Tokenization: Seperate the text into individual words

• Stemming: Get the root of the word, it helps identify and group similar words

Text data

• Term Frequency- Represents in a document the frequency of a word’s appearance

  

  • TF-Inverse Document Frequency- penalizes the words that are universal to the documents and doesn’t provide a unique quality to the documents

• Trained Embeddings- Deep learning that represents the meaning of words

Clustering

Group similar objects into groups, objects in different groups are less similar to objects in the same group

• Computers use distance, they use different distances to measure different objects

  • K-Means

    • k- predefined as in input

• Squared Euclidean Distance

• 
  

  

Classification

Pre-defined classes, requires labeled training data (supervised learning)

• More often adopted for utility

Decision Tree

• Leaves represent class labels

• Branches represent conjunctions of features that lead to those class labels

• Can provide insights on the decision process

•