Machine Learning

One of the main differences between humans and computers is that humans learn from past experiences, at least they try, but computers or machines need to be told what to do. Computers are strict logic machines with zero common sense. That means if we want them to do something, we have to provide them with detailed, step-by-step instructions on exactly what to do and that is where Machine Learning or ML comes into the picture.

What is machine learning?

Machine learning and Artificial Intelligence have emerged as some of the key technologies which have the potential to transform enterprises and the way we do business in digital era. This has further been aided by availability of on-demand computing, memory and infrastructure resources in Cloud.

Some of the key applications of machine learning within enterprises include Understanding Customer Behavior/Analytics, Streamlining and Optimizing Operations, and Generating business value through improved Customer Experience by touching each and every aspect of the business including decision making.

While there are various definitions of machine learning on the web and most of them do touch some aspect of what machine learning really is. A concise definition as per Wikipedia is as follows:

note

Wikipedia

Machine learning (ML) is the study of computer algorithms that can improve automatically through experience and by the use of data. It is seen as a part of artificial intelligence. Machine learning algorithms build a model based on sample data, known as training data, in order to make predictions or decisions without being explicitly programmed to do so. Machine learning algorithms are used in a wide variety of applications, such as in medicine, email filtering, speech recognition, and computer vision, where it is difficult or unfeasible to develop conventional algorithms to perform the needed tasks.

How do machine's learn? and How does machine learning differ from Traditional Software Systems ?

There are 2 ways machines learn.

1. A Computer programmer feeds in logic/algorithms, which are sets of rules/algorithm used to help computers perform a certain logic as predefined by the use case. Please note the logic in hand-coded in this class of learning.

In Traditional Software Systems, A Computer programmer programs the logic/algorithm as a series of logical steps/processes which is then translated into machine interpretable instructions and used to execute the programmed logic. While this paradigm of programs/algorithms have dramatically influenced every aspect of our daily life and businesses, This way of logic programming is not very effective at learning many of the tasks. For example for machines to calculate sales incentives, it may have to be programmed with every rule that would implement the logic to calcuate the sales incentive based on executive performance.

2. On the other hand, another class of algorithms help machines learn logic/rules from the data instead of hand-coding rules/logic. In this case, A Data Professional feeds in labelled/training dataset to the generalized machine learning algorithms, which inturn learn rules based on instances of data seen in labelled/training datasets.

For example, a Machine Learning based algorithm can derive the same rules by observing the historical/past data of sales performance across multiple years and infer the most optimal rules based on the data. This helps minimizing hand-coding rules as would be the case in the traditional programming paradigm.

Humans Learn from past experience and so do machines.

Various types of Machine Learning Algorithms

Machine learning algorithms are split into three main categories:

• Supervised Learning

Supervised learning is a type of machine learning where the data you put into the model is “labeled.”

Labeled simply means that the outcome of the observation (a.k.a. the row of data) is pre-known.

For example, if your model is trying to predict whether your friends will go golfing or not, You can learn that from past recordings of whether they golfed or not. You may have recorded variables such as temperature, the day of the week, hour of the week, whether it is cloudy etc. A labeled data would also have an outcome variable recorded which would reveal actual outcome of the past event. The outcome variable is marked as Yes, if your friends actually went golfing or 'No' if they did not.

• Unsupervised Learning

As you may have guessed, unsupervised learning is the opposite of supervised learning when it comes to labeled data.

With unsupervised learning, you do not know whether your friends went golfing or not — it is up to the computer to find patterns via a model to guess what happened or predict what will happen.

• Reinforcement Learning

Reinforcement learning is an area of machine learning concerned with how intelligent agents ought to take actions in an environment in order to maximize the notion of cumulative reward. For more on Reinforcement Learning, refer to ???

While the above are well known categories, There are also new categories of algorithms emerging as the field continues to evolve. For ex. Artificial General Intelligence

Due to their differences when analyzing data, these two machine learning categories are better suited for solving different problems. All forms of machine learning rely on the availability of a huge quantity of data to train algorithms.

Category	Supervised Learning	Unsupervised Learning	Reinforcement Learning
Traditional Algorithms	X-Gradient Boosting, Random Forest	Linear Regression, Support Vector Machine(SVM), Random Forest Regressor	DBScan, K nearest neighbor(KNN), Principal Component Analysis(PCA)
Deep Learning Algorithms	Multi-layer Perceptrons(MLP), Convolutional Neural Networks (CNN), Long short-term Memory (LSTM)	Multi-layer Perceptrons(MLP), Convolutional Networks (CNN)	Auto-encoders, Gaussian Mixture Model(GMM)

Another way to categorize learning algorithms is based on class of use cases. These are:

1. Classification
2. Regression
3. Clustering
4. Dimensionality Reduction
5. Decision Optimization
6. Timeseries Forecasting
7. Anomaly Detection

Use cases	Popular Algorithms
Classification	Random Forest Classifier, Decision Tree Classifier, Logistic Regression, Multilayer Perceptrons(MLP), LSTM
Regression	Random Forest Regressor, Decision Tree Regressor, Linear Regression, Multilayer perceptrons(MLP)
Clustering	K nearest neighbor, DBSCAN, K-Means
Dimensionality Reduction	Principal Component Analysis
Decision Optimization	Sequential Least Squares Programming (SLSQP), etc
Timeseries Forecasting	ARIMA(Autoregressive Integrated Moving Average),. Exponential Smoothing (ETS), Facebook Prophet, Multivariate Time-series analysis using LSTM, XGB, Light Gradient Boosting etc
Anomaly Detection	One Class SVM, Isolation Forest, Local Factor Outlier