Data Acquisition: Sources, Features and Types of Data

Hello students, welcome to Kwickprep. An AI model is only as good as the data it learns from. So where does that data come from? Today we learn data sources, types of data, features and labels, system maps, and how to collect data ethically.

First, a new term. Data acquisition means collecting the data an AI project needs. It is the second stage of the AI project cycle, right after problem scoping. We collect only data that helps solve our problem statement. And the data must be relevant, accurate, and large enough for the model to learn from.

Data sources are of two kinds, primary and secondary. Primary data is collected by you, first-hand, while secondary data was already collected by someone else. For example, a survey you run in your class is primary, while a government census report is secondary. Primary data takes more time and money to collect. But primary data fits your exact need, while secondary data may not fit perfectly.

Let us see common ways to get data. Surveys and interviews ask people questions directly. Observation and sensors record data automatically, like a temperature sensor or a traffic camera. Web scraping means a program collects data from websites, and we must follow each website's rules. Open data portals, like the Government of India site data dot gov dot in, share free datasets. An API is a way for one app to ask another app for data, like a weather app asking a weather service.

Data also comes in three types. Structured data sits neatly in rows and columns, like a marks sheet in a spreadsheet. Unstructured data has no fixed format, like photos, voice notes and videos. Semi-structured data has some tags but no strict table. An email is one example, with a To field, a Subject and free text.

Pause and predict the type of each item. A WhatsApp voice note is unstructured, because audio has no rows or columns. A railway reservation chart in Excel is structured. A product page has tagged fields like name and price, but the reviews are free text, so it is semi-structured.

Now two important words in a dataset. A feature is an input column the model uses to decide, here study hours and attendance. A label is the answer column the model must learn to predict, here the result. In the first row, five hours and ninety two percent attendance gave Pass. In the second row, one hour and sixty percent gave Fail. The third row, four hours and eighty five percent, gave Pass. The last row, zero hours and forty five percent, gave Fail.

Choosing features is a skill. Pick features that truly affect the answer, like study hours for exam results. Drop features that do not matter, like roll number, because they only confuse the model. And remove personal data the problem does not need, like phone numbers.

A system map is a drawing that shows the elements of a problem and how they affect each other. Here each box names an element, and its plus or minus sign shows the effect it feels. More rainfall raises the lake water level, so that box carries a plus. A higher lake level means more water supply to homes, again a plus. More supply means less shortage, so the shortage box carries a minus. Finally, more shortage means more demand for water tankers, a plus.

Why do we draw a system map? It shows which elements of the problem are linked. A plus sign means both rise together, and a minus sign means when one rises, the other falls. Most useful of all, it tells us which data to collect, like rainfall and lake levels.

Finally, data must be collected ethically, which means in a fair and right way. Consent means asking people for permission before collecting their data. Privacy means collecting only what you need and keeping it safe. Fairness means including every group, so the model is not biased, which means unfair to some people. Honesty means telling people how their data will be used. And in India, personal data is protected by the Digital Personal Data Protection Act of twenty twenty three, called the DPDP Act.

Let us revise what we learned today. Primary data is collected first-hand, and secondary data was already collected by someone else. Data can be structured, unstructured or semi-structured. Features are the inputs, and the label is the answer the model predicts. A system map shows how elements are linked, with plus and minus signs. And always collect data with consent, privacy and fairness.