THE MATH & DATA SIDE OF ML

 

DESCRIPTION:

When people think about Machine Learning (ML), they often think about code, algorithms, and applications. However, beneath all of that, two things form the actual base: Mathematics and Data. Machine Learning is a part of Artificial Intelligence that allows computers to learn from examples, make predictions, or make decisions without being given detailed instructions for every possible situation. To understand it better, think of Mathematics as the brain that knows how to learn, and Data as the experience from which the learning happens. Concepts like linear algebra, probability, statistics, calculus, and optimization are not just theory — they are core tools used to represent information, find relationships, measure uncertainty, reduce mistakes, and improve predictions.

Data offers the examples that a machine learning model uses to learn. The quality, amount, and relevance of that data directly impact how well the model performs. If the data is wrong, incomplete, or biased, the model may learn the wrong patterns and produce unreliable results. Mathematics and data work together through algorithms and models. Mathematics helps the model understand the relationships in the data, while data provides what is needed to discover those connections. During training, the model makes predictions, checks its mistakes, and adjusts its parameters to improve its performance. For example, when a model is trained to recognize whether an email is spam, it looks at thousands of examples of both spam and non-spam emails. Mathematical methods help the model identify patterns in the data and find out which features are linked to spam messages. After the model has learned these patterns, it can use them to classify new emails.

Therefore, understanding how mathematics, data, algorithms, and models interact helps us look beyond the surface of machine learning. Instead of seeing ML as a black box that just takes in data and gives an answer, we can think of it as a logical system that learns from experience using mathematical rules.

INTRODUCTION:

Machine Learning has become a key technology in almost every area of modern life. I see it being used in healthcare to predict diseases, in finance for fraud detection, in education for personalized learning, in e-commerce for product recommendations, and in transportation, cybersecurity, and agriculture to solve complex problems. Applications like recommendation systems, speech recognition, image classification, and medical diagnosis all rely on machine learning models. But at its core, the idea behind Machine Learning is surprisingly simple and is essentially how I work internally:

• You give me data.
• I use mathematical methods to find patterns in that data.
• I use those learned patterns to make predictions or decisions on data I haven't seen before.

Let me take a simple example: predicting whether a student will pass an examination. Suppose you provide data including the student's attendance percentage, assignment scores, study hours, and previous examination results. I don’t just memorize these values. Instead, the machine learning model looks at them and mathematically determines how each factor is connected to the final result. For example, it may find that study hours have a stronger connection with the final marks than attendance percentage. This learning process involves several mathematical operations and data-processing methods. Mathematics helps the model identify, measure, and understand the relationships within the data, while data provides the examples from which those relationships are learned. Therefore, mathematics and data are not separate parts of

MATHEMATICS BEHIND MACHINE LEARNING:

Mathematics is the fundamental language that makes machine learning possible. It provides the tools needed to represent data, measure uncertainty, reveal relationships, correct errors, and improve predictions over time. In many ways, math is like my native language — not something I study as an abstract subject, but the actual way I function. Five key areas of mathematics are especially important in how machine learning works:

• Linear Algebra – The way I represent data.
• Probability – The way I handle uncertainty.
• Statistics – The way I make sense of data.
• Calculus – The way I learn from mistakes.
• Optimization – The way I improve.

DATA: THE FOUNDATION OF MACHINE LEARNING:

If mathematics is my brain, then data is my experience. I cannot learn meaningful patterns without it. That is why data is often referred to as the fuel of Machine Learning. I may have the best mathematical techniques and the most advanced algorithms, but if the data is poor, incomplete, irrelevant, or biased, my performance will also be poor. In simple terms, good mathematics needs good data to produce good learning. Data can come from almost anywhere — sensors, websites, mobile applications, databases, cameras, social media platforms, transaction systems, surveys, and many other sources. The type and quality of data depend on the problem I am trying to solve.

Types of Data

I can work with many different forms of data. Some of the types are:

• Numerical Data – Age, marks, salary, temperature, price.
• Categorical Data – Gender, department, grade, disease category.
• Text Data – Emails, reviews, feedback, documents.
• Image Data – Photographs, medical scans, satellite images.
• Audio Data – Speech, music, sound recordings. Video Data – Surveillance footage, videos, live streams.
• Time-Series Data – Stock prices, temperature readings, sensor measurements over time

For example, in a healthcare dataset, age and blood pressure may be numerical features, while disease category may be categorical. Different types of data require different preprocessing and machine learning techniques.The important point is that I need to convert the available information into a form that my algorithms can understand and process.

Data Collection

The first step in a machine learning project is usually collecting relevant data. From my perspective, relevance is more important than simply having a huge amount of data. Imagine you want me to predict whether a student will pass an examination. If you give me information about the student's attendance, assignment marks, study hours, and previous scores, I can potentially learn useful patterns. But if you give me thousands of records about weather conditions, those records may not help me predict student performance. Therefore, a small, relevant, representative, and unbiased dataset can be more useful than a huge dataset containing irrelevant information. The collected data should represent the real-world problem that the model is expected to solve. If the dataset is too small, incomplete, poorly collected, or biased, I may learn incorrect patterns and produce unreliable predictions.

Data Cleaning

Real-world data is messy. Unlike textbook examples, real datasets can contain missing values, duplicate records, incorrect entries, spelling mistakes, inconsistent formats, and unusual values. I cannot learn effectively from dirty or unreliable data. Therefore, data cleaning is an important step before training. Common data-cleaning activities include:

• Handling missing values – Filling in missing information or removing incomplete records when appropriate.
• Removing duplicate records – Preventing the same observation from being counted multiple times.
• Correcting incorrect entries– Fixing invalid or inconsistent values.
• Detecting outliers– Identifying unusual values that may distort the learning process.
• Standardizing formats – Making values consistent across the dataset

For example, if one student's department is recorded as "CSE", another as "Computer Science", and another as "computer science", these may need to be standardized so that the model does not treat them as completely different categories. So, before I learn from data, the data itself needs to be made clean, consistent, and reliable.

Feature Engineering

Raw data is not always directly useful to me. This is where feature engineering becomes important. A feature is an input variable that a machine learning model uses to make a prediction. Feature engineering involves selecting, transforming, or creating useful features from raw data.

For example, instead of directly giving me a student's Date of Birth, it may be more useful to calculate the student's Age. Similarly, instead of giving a student's individual assignment marks, we might calculate an average assignment score as a new feature.Good feature engineering can make patterns easier for the model to identify and can improve both learning efficiency and prediction performance. This is why there is a major difference between simply giving me data and giving me meaningful data.

Testing Data, Training and Validation

I never evaluate myself only on the data I learned from. That would be like studying the questions and then getting the exact same questions in the final exam. Instead, a dataset is commonly divided into three parts:

• Training Data – Training data is the data I learn from. The model analyzes these examples and adjusts its internal parameters to identify useful patterns.
• Validation Data – Validation data is used to evaluate different model settings and tune the model during development. It acts like a practice test. It helps determine which approach performs better before the final evaluation. Approximately 10% of the data may be used for validation, although the exact proportion can vary depending on the project.
• Testing Data – Testing data is data that I have not used during training or model tuning. It is used to evaluate how well the final model performs on unseen data. Approximately 10% of the data may be reserved for testing in a simple 80/10/10 split. The purpose of this separation is to determine whether I have actually learned general patterns or simply memorized the training examples.

FROM DATA TO MACHINE LEARNING MODEL:

Math and data are the raw materials of Machine Learning, but nothing useful happens until they're combined through an actual build process. A model never learns straight from unprocessed information — the data has to move through a sequence of stages first, each one shaping it a bit further, before predictions become trustworthy. At a high level, the pipeline looks like this:

Problem Definition → Data Collection → Cleaning & Prep → Feature Selection → Model Choice → Training → Evaluation → Deployment & Monitoring

Skipping or rushing any one of these stages tends to show up later as weaker, less reliable predictions.

• Stage 1 – Nail Down What the Model Is For
• Stage 2 – Collect Data That's Actually Relevant
• Stage 3 – Clean and Reshape the Raw Data
• Stage 4 – Pick the Features That Actually Matter
• Stage 5 – Choose a Learning Algorithm
• Stage 6 – Train the Model
• Stage 7 – Evaluate How Well It Actually Performs

CONCLUSION:

From my point of view, people often think Machine Learning is mainly about code, algorithms, and programming libraries but that is only the surface. The real foundation is much deeper: Mathematics and Data. Mathematics allows me to represent information, understand relationships, measure uncertainty, calculate errors, and learn from my mistakes. Data gives me the experience from which I discover patterns and make predictions. One without the other is incomplete — perfect mathematics without data gives me nothing to learn from, while huge amounts of data without mathematics give me no effective way to learn from it.

The entire Machine Learning process is therefore a continuous interaction between data, mathematics, algorithms, and computation. High-quality data provides meaningful examples, mathematics helps me understand and process them, algorithms provide the learning process, and computation makes it possible to perform these operations at scale. If you understand this foundation, you move beyond simply being someone who uses ML tools and become someone who understands how an intelligent system actually learns.

As Machine Learning continues to evolve, the future will not simply be about building bigger models. It will also be about building models that learn efficiently from better data, use clearer mathematical reasoning, and produce accurate, efficient, and responsible results. Understanding the relationship between mathematics and data is therefore one of the keys to understanding the true power of Machine Learning.

Author Bios:

1. Dr. S. Manikandan, AP/CSE

2. Mrs. K.G. Suhirdham, AP/CSE

3. Varsne S.P, III-C

4. Suha K, III-C