Introduction to Machine Learning

Machine learning (ML) allows computers to learn and make decisions

without being explicitly programmed. It involves feeding data into
algorithms to identify patterns and make predictions on new data. It is
used in various applications like image recognition, speech processing,
language translation, recommender systems, etc. In this article, we will
see more about ML and its core concepts.

Why do we need Machine Learning?

1. Solving Complex Business Problems

2. Handling Large Volumes of Data
3. Automate Repetitive Tasks
4. Personalized User Experience
5. Self-Improvement in Performance

Types of Machine Learning

There are three main types of machine learning which are as follows:

1. Supervised learning
Supervised learning trains a model using labeled data where each input
has a known correct output. The model learns by comparing its
predictions with these correct answers and improves over time. It is used
for both classification and regression problems.
Example: Consider the following data regarding patients entering a clinic.
The data consists of the gender and age of the patients and each patient
is labeled as "healthy" or "sick".

Gender Age Label

M 48 sick

M 67 sick

F 53 healthy

M 49 sick

F 32 healthy

M 34 healthy

M 21 healthy

In this example, supervised learning is to use this labelled data to train a

model that can predict the label ("healthy" or "sick") for new patients
based on their gender and age. For example if a new patient i.e Male with
50 years old visits the clinic, model can classify whether the patient is
"healthy" or "sick" based on the patterns it learned during training.
 Supervised learning is categorized into two types −

1. Classification
Classification uses algorithms to predict categorical values, such as
determining whether an email is spam or not or whether it is true or false.
The algorithm learns to map each input to its corresponding output label.
Some common algorithms include K-Nearest Neighbors, Random forests and
Decision trees.

2. Regression
Regression is a statistical approach to analyze the relationship between data
points. It can be used to forecast house prices based on features like location
and size or estimate future sales. Some common algorithms include linear
regression, polynomial regression, and logistic regression.

Advantages
 High Accuracy:

Often achieves high accuracy due to the explicit guidance of labeled data, making
predictions precise.

 Clearer Evaluation:

Results are easier to evaluate and validate using metrics like accuracy and
precision because of known correct outputs.

 Straightforward Approach:
Offers a more direct approach for classification and regression tasks like spam
detection and weather forecasting.

Disadvantages
 Expensive Data Labeling:

Requires a large amount of pre-labeled data, which can be costly and time-
consuming to obtain.

 Prone to Overfitting:

Models can become overly tailored to the training data, leading to poor
performance on new data.

 Limited to Labeled Data:

Its applicability is restricted to scenarios where labeled datasets are available.
2. Unsupervised learning:

Unsupervised learning works with unlabelled data where no correct

answers or categories are provided. The model's job is to find the data,
hidden patterns, similarities or groups on its own. This is useful in
scenarios where labelling data is difficult or impossible. Common
applications are clustering and association.
Example: Consider the following data regarding patients. The dataset has
a unlabelled data where only the gender and age of the patients are
available with no health status labels.
Gender Age

M 48

M 67

F 53

M 49

F 34

M 21

Here unsupervised learning looks for patterns or groups within the data on
its own. For example it might cluster patients by age or gender and
grouping them into categories like "younger healthy patients" or "older
patients" without knowing their health status.

Example:
- Imagine you have a large set of images of different
animals, but none of the images are labelled (no “dog,”
 “cat,” etc.). An unsupervised learning algorithm could
group similar images together, forming clusters, even if it
doesn’t know what the animals are.

Simple Analogy:
It’s like a person sorting different types of fruits without
knowing their names. The person groups similar-looking
fruits together (apples in one group, oranges in another)
without knowing exactly what they are.

Unsupervised learning is categorized into different types.

1. Clustering
A clustering is used to group similar data points together. Clustering
algorithms work by repeatedly moving data points closer to the centre of
their group (cluster) and farther from points in other groups. This helps the
algorithm to create clear and meaningful clusters. Popular clustering
algorithm is K-means clustering.
2. Association
This is another type of unsupervised learning that uses pre-defined rules to
group data points into a cluster. It is commonly used in Market Basket
Analysis, and the main algorithm behind this task is Apriori Algorithm.

Advantages
 Handles Unlabelled Data: Can process large amounts of raw, unlabelled data
where labelling isn't feasible or available.

 Discovery of Patterns: Excellent for exploratory data analysis, allowing the

discovery of previously unknown patterns and relationships.

 Flexibility and Adaptability: Models can adapt to new data and are more flexible in
their applications.

Disadvantages
 Subjective Results:

The lack of labels makes results subjective and harder to validate or interpret
without human intervention.

 Difficult Validation:

There's no ground truth, making it challenging to quantitatively measure the

accuracy or effectiveness of the model's output.

 Susceptible to Noise:
Models may over fit to noise or spurious patterns in the data, which can lead to
inaccurate patterns.

3. Reinforcement Learning
Reinforcement Learning (RL) trains an agent to make decisions by
interacting with an environment. Instead of being told the correct answers,
agent learns by trial and error method and gets rewards for good actions
and penalties for bad ones. Over time it develops a strategy to maximize
rewards and achieve goals. This approach is good for problems having
sequential decision making such as robotics, gaming and autonomous
systems.

Example:
- In a video game, an RL agent learns to play by making
moves and receiving points (rewards) or losing lives
(penalties). Over time, it figures out the best strategies to
win the game.

Simple Analogy:
Imagine teaching a pet a new trick. Every time the pet
performs the trick correctly, you give it a treat (reward). If
the pet does something wrong, you don’t give a treat
 (penalty). Over time, the pet learns to perform the trick
correctly to get the treat.

Example: While Identifying a Fruit, system receives an input for example

an apple and initially makes an incorrect prediction like "It's a mango".
Feedback is provided to correct the error "Wrong! It's an apple" and the
system updates its model based on this feedback.
Over time it learns to respond correctly that "It's an apple" when getting
similar inputs and also improves accuracy.

Supervised or Unsupervised Learning - Which to Choose?

Choosing the right approach is crucial and will also determine the efficiency
of the outcome. To decide on which learning approach is best, the following
things should be considered −

 Dataset − Evaluate the data, whether it is labelled or unlabelled. You will

also need to assess whether you have the time, resources and expertise to
support labeling.
 Goals − It is also important to define the problem you are trying to solve and
the solution you are trying to opt for. It might be classification, discovering
new patterns or insights in the data or creating a predictive model.
 Algorithm − Review the algorithm by making sure that it matches required
dimensions, such as attributes and number of features. Also, evaluate if the
algorithm can support the volume of the data.
 Advantages
 Handles Complex Environments:

RL excels in dynamic and complex environments, such as robotics, autonomous

vehicles, and gaming, optimizing actions for long-term goals.

 No Labeled Data Required:

RL agents learn by interacting with their environment and receiving feedback

(rewards or penalties), eliminating the need for large, costly, and labeled datasets
that are often impractical to acquire.

 Continuous Improvement:

Through interaction, the agent's strategies are continuously improved, allowing it to

adapt to changing conditions and outperform static methods.

 Adaptability:

RL algorithms can adapt in real-time to dynamic environments, making them

suitable for applications where conditions are constantly shifting.

Disadvantages
 High Data and Computational Demands:

Training RL models requires vast amounts of data and significant computational

power, making the process time-consuming and expensive, especially for smaller
projects.

 Reward Function Dependence:

The entire performance of an RL agent depends on a well-designed reward

function. A poorly designed reward can lead to the agent learning the wrong
behaviors or failing to achieve the desired task.

 Complexity in Debugging and Interpretation:

It can be difficult to understand why an RL model makes certain decisions, making

it challenging to diagnose and fix problems or to interpret the agent's behavior.

 Not Ideal for Simple Problems:

RL is often considered overkill for straightforward, simple tasks, where simpler,

more efficient methods might be more appropriate.
Machine Learning Lifecycle

Below are the key steps of the ML lifecycle:

Step 1: Problem Definition

The first step is identifying and clearly defining the business problem. A
well-framed problem provides the foundation for the entire lifecycle.
Important things like project objectives, desired outcomes and the scope
of the task are carefully designed during this stage.
 Collaborate with stakeholders to understand business goals
 Define project objectives, scope and success criteria
 Ensure clarity in desired outcomes

Step 2: Data Collection

Data Collection phase involves systematic collection of datasets that can

be used as raw data to train model. The quality and variety of data directly
affect the model’s performance.
Here are some basic features of Data Collection:
 Relevance: Collect data should be relevant to the defined problem and
include necessary features.
 Quality: Ensure data quality by considering factors like accuracy and
ethical use.
 Quantity: Gather sufficient data volume to train a robust model.
 Diversity: Include diverse datasets to capture a broad range of
scenarios and patterns.

Step 3: Data Cleaning and Preprocessing

Raw data is often messy and unstructured and if we use this data directly
to train then it can lead to poor accuracy. We need to do data cleaning
and preprocessing which often involves:
 Data Cleaning: Address issues such as missing values, outliers and
inconsistencies in the data.
 Data Preprocessing: Standardize formats, scale values and encode
categorical variables for consistency.
 Data Quality: Ensure that the data is well-organized and prepared for
meaningful analysis.

Step 4: Exploratory Data Analysis (EDA)

To find patterns and characteristics hidden in the data Exploratory Data

Analysis (EDA) is used to uncover insights and understand the dataset's
structure. During EDA patterns, trends and insights are provided which
may not be visible by naked eyes. This valuable insight can be used to
make informed decision.
Here are the basic features of Exploratory Data Analysis:
 Exploration: Use statistical and visual tools to explore patterns in
data.
 Patterns and Trends: Identify underlying patterns, trends and
potential challenges within the dataset.
 Insights: Gain valuable insights for informed decisions making in later
stages.
 Decision Making: Use EDA for feature engineering and model
selection.

Step 5: Feature Engineering and Selection

Feature engineering and selection is a transformative process that involve

selecting only relevant features to enhance model efficiency and
prediction while reducing complexity.
Here are the basic features of Feature Engineering and Selection:
 Feature Engineering: Create new features or transform existing ones
to capture better patterns and relationships.
 Feature Selection: Identify subset of features that most significantly
impact the model's performance.
 Domain Expertise: Use domain knowledge to engineer features that
contribute meaningfully for prediction.
 Optimization: Balance set of features for accuracy while minimizing
computational complexity.

Step 6: Model Selection

For a good machine learning model, model selection is a very important

part as we need to find model that aligns with our defined problem, nature
of the data, complexity of problem and the desired outcomes.
Here are the basic features of Model Selection:
 Complexity: Consider the complexity of the problem and the nature of
the data when choosing a model.
 Decision Factors: Evaluate factors like performance, interpretability
and scalability when selecting a model.
 Experimentation: Experiment with different models to find the best fit
for the problem.

Step 7: Model Training

With the selected model the machine learning lifecycle moves to model
training process. This process involves exposing model to historical data
allowing it to learn patterns, relationships and dependencies within the
dataset.
Here are the basic features of Model Training:
 Iterative Process: Train the model iteratively, adjusting parameters to
minimize errors and enhance accuracy.
 Optimization: Fine-tune model to optimize its predictive capabilities.
 Validation: Rigorously train model to ensure accuracy to new unseen
data.

Step 8: Model Evaluation and Tuning

Model evaluation involves rigorous testing against validation or test

datasets to test accuracy of model on new unseen data. It provides
insights into model's strengths and weaknesses. If the model fails to
acheive desired performance levels we may need to tune model again
and adjust its hyperparameters to enhance predictive accuracy.
Here are the basic features of Model Evaluation and Tuning:
 Evaluation Metrics: Use metrics like accuracy, precision, recall and
F1 score to evaluate model performance.
 Strengths and Weaknesses: Identify the strengths and weaknesses
of the model through rigorous testing.
 Iterative Improvement: Initiate model tuning to adjust
hyperparameters and enhance predictive accuracy.
 Model Robustness: Iterative tuning to achieve desired levels of model
robustness and reliability.

Step 9: Model Deployment

Now model is ready for deployment for real-world application. It involves

integrating the predictive model with existing systems allowing business to
use this for informed decision-making.
Here are the basic features of Model Deployment:
 Integrate with existing systems
 Enable decision-making using predictions
 Ensure deployment scalability and security
 Provide APIs or pipelines for production use
Step 10: Model Monitoring and Maintenance
After Deployment models must be monitored to ensure they perform well
over time. Regular tracking helps detect data drift, accuracy drops or
changing patterns and retraining may be needed to keep the model
reliable in real-world use.

Here are the basic features of Model Monitoring and Maintenance:

 Track model performance over time
 Detect data drift or concept drift
 Update and retrain the model when accuracy drops
 Maintain logs and alerts for real-time issues

Each step is essential for building a successful machine learning model

that can provide valuable insights and predictions.
 ARTIFICAIL NEURAL NETWORK

An Artificial Neural Network (ANN) is a machine learning model inspired by the human
brain's neural structure, consisting of interconnected nodes ("neurons") organized in layers
that process information. During training, ANNs adjust the strength (weights) of connections
between neurons to learn patterns from data, enabling them to perform tasks like image
recognition, pattern association, and data classification with increasing accuracy over time.

Neural networks are among the most influential algorithms in modern machine
learning and artificial intelligence (AI). They underpin breakthroughs in computer
vision, natural language processing (NLP), speech recognition and countless real-world
applications ranging from forecasting to facial recognition.
 How do neural networks work?
Neural networks are composed of a collection of nodes. The nodes are
spread out across at least three layers. The three layers are:

Layers
1. Input Layer: This is where the network receives its input data. Each
input neuron in the layer corresponds to a feature in the input data.
2. Hidden Layers: These layers perform most of the computational
heavy lifting. A neural network can have one or multiple hidden layers.
Each layer consists of units (neurons) that transform the inputs into
something that the output layer can use.

3. Output Layer: The final layer produces the output of the model. The
format of these outputs varies depending on the specific task like
classification, regression.

Working of Neural Networks

Forward Propagation (Forward Pass):

During forward propagation, also known as the forward

pass, input data is fed into the neural network, and
computations are performed layer by layer until the output
is generated. Each neuron in the network receives input
signals, applies a specific activation function to them, and
passes the result to neurons in the subsequent layer. This
process continues until the final output is produced.
Forward propagation is responsible for making predictions
based on the current parameters (weights and biases) of
the network.

Calculation of the Loss Function:

Once the output is generated, the next step is to evaluate

the performance of the model by comparing its predictions
with the actual target values from the training data. This is
done using a loss function, also known as a cost function or
objective function, which quantifies the difference between
the predicted outputs and the ground truth. Common loss
functions include mean squared error (MSE), cross-
entropy loss, and hinge loss. The goal during training is to
minimize this loss function, thereby improving the model’s
accuracy.

Backward Propagation (Backward Pass or Backpropagation):

After computing the loss function, the network adjusts its

internal parameters, specifically the weights and biases, to
minimize the loss and improve performance. Backward
propagation, also known as backpropagation, is the
process of updating these parameters by calculating the
gradient of the loss function with respect to each
parameter. This involves propagating the error backward
through the network, layer by layer, using techniques such
as the chain rule of calculus. The gradients are then used
to update the parameters using optimization algorithms
like gradient descent or its variants. By iteratively
adjusting the parameters based on the computed
gradients, the network gradually learns to make more
accurate predictions and minimize errors.

Overall, these three components—forward propagation,

calculation of the loss function, and backward propagation
—are essential steps in the training of artificial neural
networks. Together, they enable the network to learn from
data, optimize its parameters, and improve its performance
over time, ultimately making it capable of solving complex
tasks and making accurate predictions.

Neural network training

Just like other machine learning algorithms, a neural net requires rigorous
training to perform well on testing. To train a network, a single neuron
computes:

z=∑i=1nwixi+b

a=σ(z)

Where:

 xi = input feature,
 wi = weight,
 b = bias,
 z = weighted sum (linear transformation),
 σ = activation function (nonlinear transformation),
 a = output,
 Iteration
This process of forward propagation, loss calculation, backpropagation
and weight update is repeated for many iterations over the dataset. Over
time, this iterative process reduces the loss and the network's predictions
become more accurate.
Through these steps, neural networks can adapt their parameters to
better approximate the relationships in the data, thereby improving their
performance on tasks such as classification, regression or any other
predictive modelling.

Types of Neural Networks

There are seven types of neural networks that can be used.

 Feedforward Networks: It is a simple artificial neural network

architecture in which data moves from input to output in a single
direction.

 Singlelayer Perceptron: It has one layer and it applies weights, sums

inputs and uses activation to produce output.
 Multilayer Perceptron (MLP): It is a type of feedforward neural
network with three or more layers, including an input layer, one or more
hidden layers and an output layer. It uses nonlinear activation
functions.

 Convolutional Neural Network (CNN) : It is designed for image

processing. It uses convolutional layers to automatically learn features
from input images, enabling effective image recognition and
classification.

 Recurrent Neural Network (RNN) : Handles sequential data using

feedback loops to retain context over time.

 Long Short-Term Memory (LSTM) : A type of RNN with memory cells

and gates to handle long-term dependencies and avoid vanishing
gradients.

Advantages

Neural networks are widely used in many different applications because of

their many benefits:
 Adaptability: Neural networks are useful for activities where the link
between inputs and outputs is complex or not well defined because
they can adapt to new situations and learn from data.
 Pattern Recognition: Their proficiency in pattern recognition renders
them efficacious in tasks like as audio and image identification, natural
language processing and other intricate data patterns.
 Parallel Processing: Because neural networks are capable of parallel
processing by nature, they can process numerous jobs at once which
speeds up and improves the efficiency of computations.
 Non-Linearity: Neural networks are able to model and comprehend
complicated relationships in data by virtue of the non-linear activation
functions found in neurons which overcome the drawbacks of linear
models.

Disadvantages

Neural networks while powerful, are not without drawbacks and

difficulties:
 Computational Intensity: Large neural network training can be a
laborious and computationally demanding process that demands a lot
of computing power.
 Black box Nature: As "black box" models, neural networks pose a
problem in important applications since it is difficult to understand how
they make decisions.
 Overfitting: Overfitting is a phenomenon in which neural networks
commit training material to memory rather than identifying patterns in
the data. Although regularization approaches help to alleviate this, the
problem still exists.
 Need for Large datasets: For efficient training, neural networks
frequently need sizable, labeled datasets; otherwise, their performance
may suffer from incomplete or skewed data.

Applications

Neural networks have numerous applications across various fields:

1. Image and Video Recognition: CNNs are extensively used in

applications such as facial recognition, autonomous driving and
medical image analysis.

2. Natural Language Processing (NLP): RNNs and transformers power

language translation, chatbots and sentiment analysis.

3. Finance: Predicting stock prices, fraud detection and risk

management.

4. Healthcare: Neural networks assist in diagnosing diseases, analyzing

medical images and personalizing treatment plans.

5. Gaming and Autonomous Systems: Neural networks enable real-

time decision-making, enhancing user experience in video games and
enabling autonomous systems like self-driving cars.