Machine Learning​

Algorithm & business use cases​

Share on linkedin
Share on facebook
Share on twitter
Share on email

At this moment Machine learning is the single most important thing in Artificial Intelligence as it provides tremendous value to businesses across industries.  Even though the concept of ML has already been here for almost half a century, it is only now average businesses are able to commercialize it and drive value from it because of the increase of processing speed and the huge amounts of data we generate everyday.  

Netflix, the streaming company, showed that by leveraging ML to provide better recommendation on the platform they are able to generate extra 1 billion dollar revenue from increasing customers churn.  It’s not just Netflix.  Most businesses are actively investing and experimenting in ML to extract value from data they’ve been collecting.  In 2017, Baidu and Google both invested more than 20 billions into AI  and McKinsey even suggest that ML has the potential to generate 9.5 trillion dollar of value in 19 industries.  

What is Machine Learning?

ML, just like what the phrase implies, is a machine that can learn.  Unlike traditional program that follows a series of rules and policy, ML are able to conduct deduction and improve itself upon the data they were fed.  When new data were fed into the model, they can update their understanding of the original problems based on it.  

Let’s say there is a person who has no idea what beauty is and you show him different people and tell him whether each of this person is beautiful or not.  After he sees more and more people, that is the data set becomes larger and larger, he will have a better understanding of what beauty is. 

A few things to note with the above example is that the data may not be clean or it may even be biased or the dataset is not big enough for the learner to conclude what is beauty so it’s really important to have a large dataset that are both clean and not biased. 

Let’s say there is a person who has no idea what beauty is and you show him different people and tell him whether each of this person is beautiful or not.  After he sees more and more people, that is the data set becomes larger and larger, he will have a better understanding of what beauty is. 

A few things to note with the above example is that the data may not be clean or it may even be biased or the dataset is not big enough for the learner to conclude what is beauty so it’s really important to have a large dataset that are both clean and not biased. 

Machine Learning Definition

Supervised Learning

Supervised learning uses labeled data to train the model just like the above example with the person who has no idea what beauty is.  You can also think of it as a teacher teaching a student by telling him every single answer to each question.  As the student is fed with more answers to questions, he will build deeper and deeper understanding to the same type of problems.

A labeled dataset means each data has an input and an associated output.  A picture of a number is the input and what number the picture represents is the output.  Let’s say we are to build a model that can recognize number 0~9 with supervised learning.  First we have to feed the model with a bunch of pictures of a number and tell the model what each number represents.  After the model processed all the data, if we are lucky, it will be able to recognize picture that it wasn’t fed before and successfully identify which number it represents. 

In the real world, supervised learning is the most used ML technique driven by its simplistic nature and the fact that we have too much data now.  Consumer companies can train model based on purchase history of customers and build recommendation system from it or manufacturers can implement IoT in their assembly line and train the model to detect defects in the production line.  

Algorithms and its use cases

Linear Regressoin

 It performs a regression task. Regression models a target prediction value based on independent variables. It is mostly used for finding out the relationship between variables and forecasting.  Regression techniques may differ based on the number of independent variables and the type of relationship between the independent and dependent variables.

  • Assess opportunities and risks like demand analysis and conversion analysis
  • Optimize operation efficiency and effectiveness by bringing data driven decision making into the organization culture
  • Predicting price elasticity, optimize pricing strategy and assess market dynamic
  • Discover consumer insight through regression analysis

Logistic Regression​

Logistic regression is a classification algorithm used to assign observations to a discrete set of classes like zero or one, cat or dog, or anything classified unlike linear regression which outputs continuous number values.  But just like linear regression assumes that the data follows a linear function, Logistic regression models the data using the sigmoid function.

  • Classify customer group based on their probability of repaying loan
  • Determine whether or not a transaction is fraud based on pattern and behavior
  • Predicting whether or not a tumor is benign
  • Through journey analytics and customer behavior to determine whether or not a customer will convert

Naive Bayes​

Naive Bayes is a probabilistic algorithm that’s typically used for classification problems like a spam filter or a recommendation system.  It is a collection of classification algorithms based on Bayes’ Theorem. It is not a single algorithm but a family of algorithms where every pair of features being classified is independent of each other.

  • Assess market response to a product with semantic analysis on social media
  • Building a spam filter for email
  • Building a recommender system with data collected with data mining
  • Discover new market insight through regression analysis

Decision Tree​

Decision Tree Analysis is a predictive modelling tool that are widely used in statistic, data mining, and machine learning.   decision trees are constructed with algorithm that identifies way to split data based on different conditions.  It is one of the most widely used and practical methods for supervised learning.  It can be used for both classification and regression tasks. 

  • Provide framework for making decision - make organization more data driven
  • Uncover consumer insight to really understand what makes a customer tick
  • Help assessing risks and feasibility of alternative

Random Forest​

Random forest is a supervised learning algorithm for classification, regression and other tasks that operates by constructing multiple decision trees at training time.  It will return the class that is the mode of the classes of the individual trees if it were a classification task or mean prediction of the individual trees if it were regression tasks

  • Prediction electric usage of a grid
  • Optimize efficiency of an infrastructure project
  • Assess quality of the product prior to production

Support Vector Machine​

SVM is a supervised ML algorithm which can be used for both classification and regression.  However, it is mostly used in classification problems.  It constructs a hyperplane or a set of hyperplanes in a high or infinite-dimensional space.

  • Determining where the face is in a photo or video
  • Predicting conversion rate of advertising, website or other channels
  • Recognizing font used

AdaBoost​

AdaBoost, short for “Adaptive Boosting”, is the first practical boosting algorithm that are used to solve real world problems.  There is this saying among data scientist that when nothing works, boosting does.  While Adaboost is just one of the boosting algorithm, it is however the best starting point for understanding boosting and is a more general solution among boosting algorithms. 

  • Determine whether or not a transaction is fraudulent
  • Recognizing image with a lower operating cost (compare with deep learning)
  • Predicting customer churn

Gradient Boosting Trees (GDBT)​

Gradient boosting is a supervised ML algorithm for both classification and regression tasks.  It produces a prediction model in the form of an ensemble of week prediction models produced by decision trees.  

  • Predicting demand of products
  • Adjusting price based on market dynamic
  • Provide decision framework for management

Unsupervised Learning

In the real world clean and perfectly labeled data is not always accessible and sometimes researchers are finding answers they don’t know where to look for.  This is where unsupervised learning comes in.  

Unsupervised learning eat un labeled data and categorize them based on patterns, traits, or other elements.  For example, if can have a bunch of pictures of dog without labeling what type of dog each picture represents, we can use unsupervised learning to categorize different spices of dog even though the model doesn’t know what it stands for. 

Common practices include customer journey analytics: using behaviors on a webpage to categorize different buying patterns, or finding outlier: bank using purchasing behavior to determine whether it is a fraud.

Insight related to machine learning