Basics of Machine Learning – An Overview

Machine learning
We can think of machine learning as the science of getting computers to learn automatically.
It’s a form of artificial intelligence (AI) that allows computers to act like humans, and
improve their learning as they encounter more data.
With machine learning, computers can learn to make decisions and predictions without being
directly programmed to do so. The process uses algorithms to build models that can then be
applied to a whole host of different purposes.
Machine learning was defined in 90’s by Arthur Samuel as "a field of study that gives the
ability to the computer for self-learn without being explicitly programmed”. It is mainly
focused on the development of computer programs which can teach themselves to grow and
change when exposed to new data.
IBM defines Machine Learning as “A branch of artificial intelligence (AI) and computer
science which focuses on the use of data and algorithms to imitate how humans learn,
gradually improving its accuracy.”

Algorithms
In the simplest terms, an algorithm is a set of instructions that a computer needs to follow to
complete a particular task. In relation to machine learning, algorithms analyse input data to
predict output values within an acceptable range.
As these algorithms receive new data, they ‘learn’ to optimise their processes, meaning they
improve performance and become more intelligent. As we’ll see, there are four main types
used in machine learning: supervised learning, unsupervised learning, semi-supervised
learning, and reinforcement learning.

Artificial intelligence
Artificial intelligence (AI) is a branch of computer science that focuses on developing
computers and machines that can perform tasks that usually require human intelligence. Such
software systems operate in an intentional, intelligent, and adaptive manner.
AI systems often use real-time data and inputs to respond to situations and make decisions.
They can analyse huge amounts of information in very short spaces of time. Machine
learning is just one of the subsets of artificial intelligence.

The basics of machine learning

A good place to start is with an explanation of how machine learning algorithms work. There
is a fairly famous quote that does exactly that. It comes from Tom Mitchell, an American
professor and expert in machine learning. He explains the machine learning meaning in the
following terms:
‘A computer program is said to learn from experience E with respect to some task T and
some performance measure P, if its performance on T, as measured by P, improves with
experience E.’
Let’s break that down a little more with an example. Let’s say you want a machine learning
program to predict weather patterns in a particular area (task T). You can feed past weather
pattern data (experience E) through your algorithm. If your algorithm is successful in
learning, it will be able to more accurately predict weather patterns (performance measure P).
Of course, there isn’t just one type of algorithm you can use to apply to everything. In reality,
there are thousands of highly specialised tools and programs developed for all kinds of real-
world problems.

Why we need Machine Learning:

Data is growing day by day, and it is impossible to understand all of the data with higher
speed and higher accuracy. More than 80% of the data is unstructured that is audios, videos,
photos, documents, graphs, etc. Finding patterns in data on planet earth is impossible for
human brains. The data has been very massive, the time taken to compute would increase,
and this is where Machine Learning comes into action, to help people with significant data in
minimum time.

What is machine learning used for?

You probably come into contact with machine learning algorithms on a daily basis without
realising it. A few the creative ways you might encounter the technology include:
• Automation - Perhaps the most high-profile machine learning use is in the
automation of tasks humans usually perform. The ability for a computer to think and
act without being programmed has incredible potential.
• Recommendations - Based on previous input data, machine learning can recommend
products and services that users or customers might like. This is perhaps one of the
most common forms of machine learning you’ll see in your day-to-day life.
• Insights - Machine learning algorithms can process and analyse huge sets of data.
Often used in the field of big data, such insights can help businesses understand their
customers and healthcare professionals understand their patients.
• Detection - The way that machine learning works makes it ideal for spotting
anomalies in patterns. As algorithms learn what ‘normal’ is, they become more adept
at detecting when things go wrong.
These are just a small sample of the types of areas where machine learning is being used.
Excitingly, it’s a field that’s still relatively young. As computing power increases and
algorithms become more complex, we’ll see many more uses for machine learning.
Examples of machine learning
To give a clearer picture of how machine learning is being used today, let’s explore some
real-life instances of the technology at work. Some of these machine learning examples are
ones you may have encountered directly, while others may impact you in ways you’ve never
noticed.

• Search engines - Search engines like Google use machine learning in a variety of
different ways. By watching how users respond to the results displayed when you
make a search, algorithms can refine which pages are displayed. The Google
RankBrain algorithm assesses what users might be looking for when they make a
search. Understanding this type of algorithm plays an essential part in things like
Search Engine Optimisation (SEO) and other forms of digital marketing. It also
means you get useful, relevant and high-quality results when you search online.
• Speech recognition - Virtual personal assistants have been around for a while now.
With services like Siri, Alexa, and Google Now, you can ask questions, set reminders,
and even control various elements of your home. All of these use speech recognition
and language analysis powered by machine learning. By using deep learning
algorithms and neural networks, these digital assistants can perform a whole host of
functions. Often, the more data they gather from people speaking, the more accurate
they become.
• Fraud detection - As many of our financial services move to digital platforms, the
risk of fraud and scams increases. To combat such issues, machine learning
algorithms have been devised. These programs work on large data sets to find
correlations in user behaviour that could lead to fraud. They look at wide-scale
patterns to identify anything out of the ordinary. A good example of machine learning
at work in reducing fraud is Danske Bank. Previously, their labour-intensive means of
examining fraud created 1,200 false positives a day. They were only detecting 40% of
fraud. By introducing a deep learning solution, they saved time, reduced false
positives by 60%, and increased true positives by 50%.
• Medical diagnosis - Another field that is producing massive amounts of data is that
of healthcare. Individual patients, as well as groups of people, are creating
information about diagnostics, treatments, and conditions. These big data sets can
help build predictive models on a range of illnesses and their treatments. IBM’s
Watson for Genomics, for example, uses AI and ML to allow clinicians to provide
personalised care to cancer patients. This type of precise approach to medicine can
mean more effective treatments for more people. As far as machine learning
applications go, this is one of the most valuable.
• Customer support - Compared to some of the other uses for machine learning on the
list, this might seem a little mundane. However, it’s still a great example of ML in
action. By using algorithms to assess interactions between customers and companies,
it’s possible to create things like chatbots and virtual assistants. These services
respond to queries and simulate real conversations, improving customer experience.
They can help to ensure clients receive the help they need while saving organisations
time and money. Plus, the more data the chatbot or assistant receives, the more
accurately it can help customers.
The main types of machine learning algorithms
Now that we’ve answered ‘what is machine learning?’ in basic terms, let’s take a look at
some of the different types of machine learning algorithms. Many problems that ML sets out
to solve require a bespoke approach. As such, the types of instructions needed for each will
be very different. However, there are generally three main categories that these algorithms
fall into:

1. Supervised Learning:
With this method of machine learning, you train the algorithm using a labelled set of data to
learn from. So, there are already some known answers, and it can determine whether new
data matches it. As it produces results, it can evaluate them based on information you’ve
already provided. The more data you give it initially, the more it knows about unseen data.
Here, the raw data divided into two parts. The first part is for training the algorithm, and the
other region used for test the trained algorithm.
In supervised learning, the data-driven model is built by processing a known labelled dataset
that includes desired inputs (features) and outputs (labels/responses). Supervised learning
finds the model that generates the outputs based on the inputs. A physics-driven model is a
theory-based mathematical mapping that relates the input and output, whereas supervised
learning identifies patterns in the available dataset, learns from observations, and makes the
necessary predictions based on statistical mapping of inputs and outputs.
To summarise, supervised learning uses the data patterns to predict the values of additional
data for the labels. This method will commonly use in applications where historical data
predict likely upcoming events. For example, it can anticipate when transactions are likely to
be fraudulent or which insurance customer is expected to file a claim.
Supervised Learning is mainly divided into two parts. Regression algorithms are used to
determine continuous values such as price, income, age, etc. and Classification algorithms are
used to forecast or classify the distinct values such as Real or False, Male or Female, Spam or
Not Spam, etc.
Regression:
Regression finds correlations between dependent and independent variables. Therefore,
regression algorithms help predict continuous variables such as house prices, market trends,
weather patterns, oil and gas prices etc.
The Regression algorithm’s task is finding the mapping function so we can map the input
variable of “x” to the continuous output variable of “y.”
Classification:
Classification is an algorithm that finds functions that help divide the dataset into classes
based on various parameters. When using a Classification algorithm, a computer program
categorizes the data into various categories depending on what it learned.
Classification algorithms find the mapping function to map the “x” input to “y” discrete
output. The algorithms estimate discrete values (in other words, binary values such as 0 and
1, yes and no, true or false) based on a particular set of independent variables. To put it
another, more straightforward way, classification algorithms predict an event occurrence
probability by fitting data to a logit function.
Classification algorithms are used for things like email and spam classification, predicting the
willingness of bank customers to pay their loans etc.

2. Unsupervised Learning:
Unsupervised Learning is the second type of machine learning, in which unlabelled data is
used to train the algorithm, which means it used against data that has no historical labels.
What is being showing must figure out by the algorithm. In other words, the programme is
trained with data that isn’t labelled. It doesn’t know what the data represents. Instead, the
computer detects patterns, finds rules within it, and summarises where there are relationships
in the data.
Unsupervised learning is characterized by the input of raw information directly to the
algorithm without pre-processing of the data and without knowing the output of the data. It
processes datasets to identify patterns, relations, and commonalities without using examples,
labels, and human instruction. This ML technique organizes the data in a certain way that
describes the structure, variance, density, distribution, etc. of the dataset. According to the
data segments, it makes clusters of data with new labels.
This learning technique works well on transactional data. For example, it can identify
segments of customers with similar attributes who can then be treated similarly in marketing
campaigns. Or it can find the primary qualities that separate customer segments from each
other. These algorithms are also used to segment text topics, recommend items and identify
data outliers. Unsupervised Learning can further be divided into two parts.
Clustering:
Clustering is the type of Unsupervised Learning in which similar entities of unlabelled data
are grouped together, and then the resulting grouped data is used to form clusters. The goal of
the unsupervised machine learning technique is to find similarities in the data points, group
similar data points together and to figure out accordingly which data points belong to which
cluster.
Dimensionality Reduction:
Reduction is the type of Unsupervised Learning in which the dimensions of the data are
reduced to remove any unwanted data from the input. This technique is used to remove the
undesirable features of the input data. It relates to the process of converting a set of data
points having large dimensions into the same data having a smaller size. This technique is
used while solving machine learning problems to obtain better features.

3. Reinforcement learning:
Reinforcement Learning is the third type of machine learning in which no raw data is given
as input. Instead, the reinforcement learning algorithms have to figure out the situation on
their own. Reinforcement learning is frequently used for robotics, gaming, and navigation.
With reinforcement learning, the algorithm discovers through trial and error which actions
yield the most significant rewards. This type of learning has three main components - the
agent which can be described as the learner or decision maker, the environment which is
described as everything the agent interacts with and the actions which represents what the
agent can do.
The objective is that the agent takes actions that maximise the expected reward over a given
measure of time. The agent will reach the goal much quicker by following a good policy. So
the purpose of reinforcement learning is to learn the best plan.
This method of machine learning is focused on continuous learning and reward using
unlabelled data. A useful way of thinking about this concept is with video games. If a
computer wins a game, it receives positive feedback. It then continues refining the moves it
takes to win the game to become more effective. Often, this means replaying it many times
and getting feedback on each.
Role of Machine Learning in the Petroleum Industry
Applying machine learning in oil and gas operations allows for complex problems to be
solved quickly and efficiently. Machine learning algorithms can be used for case-based
reasoning (CBR).
This means that the algorithms can be used to quickly sort through massive databases of
recorded problems. In oil and gas industries, various types of data are collected from the
surface and subsurface to understand the hydrocarbon potential. Hydrocarbon potential is a
statistical method which utilizes the geochemical data of the source rocks to calculate the
hydrocarbon generation and to quantify the petroleum resource potential.
The machine learning methods provides relationship between input variables and predicts the
output, which makes handling the immense amount of data in petroleum industries more
convenient. Given below is a table which associates petroleum extraction activities with
applications in Artificial Intelligence and Machine Learning:

The use of AI in the oil and gas industry is currently advancing rapidly, as the idea of AI
increasingly infiltrates different stages of the sector, such as intelligent drilling, intelligent
development, intelligent pipeline, and processing, so on and so forth. Developers have
created a range of realistic application technologies in research and production using
artificial intelligence algorithms.
Predominantly, in the oil and gas industry, a structure called Artificial Neural Network
(ANN) is used to process large amounts of data and to achieve its best performance. Inputs
are run through the neural network, and it will model the input data to yield the desired
output. ANN is used as an effective machine learning method to solve non-linear and
complex problems which cannot be solved by linear relationship. ANN model helps predict
pipeline conditions, by predicting failure rate and mechanical reliability. It is also applied in
drill bit diagnosis, improvement and optimization of well production and identification of
geological data at drill site.
In Reservoir Engineering, ANN is used for estimation of reservoir properties like
permeability and porosity. It also helps solves problems associated with Reservoir
Engineering including:
• Reservoir Characterization
• Process Screening
• Enhanced Oil Recovery (EOR)