Machine learning is a subset of artificial intelligence that uses algorithms and data to gradually improve accuracy. There are several types of machine learning including supervised learning, which uses labeled training data; unsupervised learning, which finds patterns without labels; and reinforcement learning, which uses trial-and-error. Machine learning has many applications such as image classification, customer segmentation, and fraud detection. It provides benefits like accuracy, automation, and scalability but also has limitations such as potential bias if the training data is biased.
Introduction to Machine Learning: Key Concepts, Algorithms, Applications and More
1. 1
1
What is Learning?
• Learning is any process by which a system improves performance
from experience. (Herbert Simon)
2. Machine learning
• ML is the subset of AI
• It focuses on the use of data and algorithms (IBM)
• to imitate the way that humans learn
• gradually improving its accuracy.
• ML applications learn from experience (i.e., data) like
humans do
8. Supervised Learning
• needs supervision
• machine is trained with well-labeled data, which means some data is already
tagged with correct outputs
• It is commonly used in applications where historical data predicts likely future
events.
• Identifying if credit card transaction is fraud or not based on historical data
9. Unsupervised Learning
• no supervision
• does not require labeled data to train a machine
• make groups of unsorted information based on some patterns and differences
• The goal is to explore the data and find some structure within
10. Semi-supervised learning
• uses both labeled and unlabeled data for training – typically a small amount of
labeled data with a large amount of unlabeled data
• is useful when the cost associated with labeling is too high to allow for a fully
labeled training process
11. Reinforcement
• is a behavioral machine learning model that is similar to supervised learning, but
the algorithm isn’t trained using sample data.
• model learns as it goes by using trial and error.
• successful outcomes will be reinforced to develop the best recommendation or
policy for a given problem
14. More ML applications
• Customer service/chatbots for facilitating communication with
businesses
• Recommender systems for customized recommendations
• Image classification e.g., auto-tagging, criminal detection, content
filtering
• Market/customer segmentation grouping customers according to
how and why they buy
• Text analysis to capture important information from documents
• Dynamic pricing like in ola, uber, airline etc.
16. Types of Machine Learning Problems
• Classification: classify the data in different classes
• 2 classes –binary classification problem
• More than 2 classes –multi-nomial classification problem
• Use data to make a prediction about a discrete set of values or categories
• Regression : need is to predict on continuous scale (numerical values)
• Predicting price of house, stock price, temperature etc. based on historical
data
17. Types of ML Problems (Contd.)
• Clustering: categories data into different groups or clusters
• Time-series forecasting: predictions based on historical time series data
• Time series data: group of observations on a single entity over time e.g. daily,
monthly, hourly etc.
• Seasonal: changes which repeat themselves within a fixed period (1 year) e.g. cost of
fruits in summer and winter, traffic in weekdays and weekends, umbrella sale
• Cyclical: changes which do not have a fixed period (several years) but are predictable
to some extent like daily variation in temperature, buying Apple phone
• Trend: captures variation over a very long time period (e.g., 20, 50, 100 years) e.g.,
increase in global temperature, increase in pollution, decrease in deaths because of
hunger etc.
• Irregular: sources of variations other than trend and cyclic. E.g. rise in prices of steel
due to strike in the factory, high demand for cement after earthquake etc.
18. Types of ML Problems (Contd.)
• Anomaly detection: identification of observations that deviate from a
dataset's normal behavior.
• E.g. ATM withdrawal pattern
• credit card fraud transactions detection
• Ranking: order the results query based on some criteria
• Recommendation: need to recommend items for possible
consumption
• Data generation: need to generate data such as images, videos,
articles, posts, etc.
19. Advantages of ML
• Accurate: More the data better the accuracy of the system
• Automated: Learns new patterns automatically
• Fast: Generates answers very fast
• Customizable: ML models are custom built to particular problem based on the
data
• Scalable: ML algorithms can scale to handle massive amounts of data
20. ML is not perfect
• ML is not based in knowledge but data
• ML models are difficult to train considering the resources required
• ML is prone to data issues like incorrect data
• ML is often biased if data is biased
21. ML vs statistics terminology
In machine learning, a target
is called a label
In statistics, a target is called
a dependent variable
A variable in statistics is
called a feature in machine
learning.
A transformation in statistics
is called feature creation in
machine learning.