Revision Notes: AI Project Cycle - Artificial Intelligence for Class 10 PDF Download

Introduction

The AI project cycle is a series of steps followed to plan, build, evaluate and deploy an AI model or solution. While it shares similarities with general IT project workflows, AI projects place special emphasis on data, model training and evaluation. A typical IT project cycle often includes the following stages:

• Design: Thinking through the interface, user experience and all functional aspects before implementation.
• Development: Building the solution by writing code and implementing the designed features.
• Testing: Checking that modules and functions work correctly and meet the requirements.
• Deployment: Releasing the application or software for real‑world use.

AI Project Cycle Stages

The AI project cycle typically emphasises stages that are specific to data and models. The main stages are:

• Problem Scoping
• Data Acquisition
• Data Exploration
• Data Modelling

Problem Scoping

Whenever we start any work, certain problems are associated with the work or process. Problems can be small or big, sometimes ignored, sometimes urgent; properly defining the problem is the first step to solving it.

Important aspects of problem scoping:

• Start every AI project by clearly scoping the problem.
• Figure out the exact problem and possible solution approaches.
• Write a clear problem statement that specifies the expected outcome and constraints.

Problem Scoping - Example

Consider this example:

• The world’s largest diamond is at risk because a person known as Mr. X has threatened to steal it.
• Mr. X has not been located, making the situation urgent.
• As the newly appointed Chief Security Officer, your mission is to strengthen the diamond’s security so it is very difficult for Mr. X to succeed.
• Using AI concepts, devise ways to improve surveillance, identification and access control around the diamond.

Three main questions to answer in this problem:

1. Identify Mr. X: Gather intelligence about his identity, background, motives and possible locations.
2. Track Mr. X: Monitor movements using surveillance, social media signals and other tracking techniques to locate him.
3. Secure the area: Put in place physical measures (guards, barriers) and AI systems (face recognition, anomaly detection) to prevent unauthorised access.

The entire process of finding a focused, workable solution is called problem scoping.

Themes for Problem Scoping

Look around and select a field where you want to solve a problem. Typical fields include:

• Agriculture
• Education
• Banking
• Health
• Security
• Infrastructure and Transportation

If we talk about agriculture, important aspects to consider are:

• Pest issues
• Yield rates
• Sowing and harvesting patterns

To understand and scope a problem effectively, use the 4Ws (Who, What, When, Where) and sometimes Why. These questions help clarify stakeholders, the intended outcome, timing and location of the problem and constraints.

The Problem Scoping Template

A problem scoping statement is prepared at the start (or end) of the scoping stage. A template typically records:

• The problem statement (what you want to solve)
• Goals and success criteria (how you will measure success)
• Stakeholders and users (who is affected)
• Constraints and risks (time, budget, privacy, legal)
• Available data sources and initial plan for data acquisition

Data Acquisition

Data Acquisition is made of two words:

• Data: Raw facts, figures or observations collected for reference or analysis.
• Acquisition: The process of obtaining those data for the project.

Note: The stage of acquiring data from relevant sources is called data acquisition.

Data Acquisition - Example (continued)

Continuing the diamond security example, suppose authorities give a list of people allowed access:

• Maintenance personnel
• Officials
• VIP visitors
• Other permitted staff

Your challenge is to ensure that no unauthorised person enters the premises. Useful data to collect could include:

• Photographs of all authorised people
• Photographs (or video) of likely unauthorised persons (if available)
• Images of the premises where the diamond is kept
• Photographs of all visitors (for training an access-control model)

Types of Data

Common types of data used in AI projects are:

1. Numerical data
2. Categorical data
3. Time series data
4. Text data

These can be described as quantitative (numbers, counts, measurements) or qualitative (labels, categories, descriptions).

Features of Data

Data features are the individual measurable properties or attributes you collect for each observation. They must be relevant to the problem statement. For example, when analysing a batsman’s weakness, useful features may include:

• Speed of the ball
• Type of dismissal
• Type of bowler
• Type of swing
• Type of spin
• Number of balls faced

Methods of Data Acquisition

The most common methods to acquire data are:

• Surveys: Collect responses via Google Forms, Microsoft Forms or similar tools.
• Web scraping: Use tools or libraries (for example, Scrapy, ParseHub, ScrapeHero, Visual Web Ripper, Import.io) to extract data from websites.
• Sensors: Devices that convert physical parameters (temperature, humidity, pressure) into electrical signals which are digitised for analysis.
• Cameras: Capture images or video for computer vision tasks.
• Observations: Collect data by watching behaviours, events or physical characteristics in natural settings.
• APIs: Application Programming Interfaces that provide structured access to data from services and platforms.

Open Source & Crowdsourcing Datasets

Some platforms and sources useful for data acquisition and labelling include:

• Lionbridge AI
• Amazon Mechanical Turk
• LabelBox
• Figure Eight
• Kaggle
• Government portals such as http://mospi.nic.in/data

Big Data for AI

Big Data refers to collections of data that are very large in volume, generated rapidly and often diverse in type, so that traditional data management tools cannot store or process them efficiently.

• Such datasets grow rapidly over time.
• They typically require specialised storage and processing tools.

Examples of Big Data

• Stock exchanges: Huge volumes of transaction and price data.
• Social media platforms: Large amounts of user-generated content and interaction data.
• Streaming services: Video content metadata, viewing histories and recommendation usage.

Types of Big Data

Big data is often described by three key characteristics known as the three Vs:

• Volume: Very large amounts of data.
• Velocity: Data generated and processed at high speed.
• Variety: Different types of data (text, numeric, images, audio).

Training, Validation and Test Sets

• Training set: The data used to train the model.
• Validation set: Data the model has not seen during training; used to tune hyperparameters and avoid overfitting.
• Test set: A final, unseen dataset used to evaluate the model’s generalisation performance after training and tuning are complete.

Data Exploration

Data exploration includes techniques and tools used to understand and visualise data, often by using summary statistics and charts. It helps reveal trends, relationships, missing values and anomalies that guide later modelling choices.

Need for Data Visualisation

• Quickly identify trends, relationships and patterns in the data.
• Decide which modelling strategies to use later.
• Communicate insights clearly to others.
• Detect data quality issues (missing values, outliers) early.

Data Visualisation Tools

• Microsoft Excel
• Tableau
• QlikView
• DataWrapper
• Google Data Studio

Modelling

• Artificial Intelligence (AI): Techniques that enable computers to perform tasks that typically require human intelligence.
• Machine Learning (ML): A subset of AI where machines improve at tasks by learning patterns from data.
• Deep Learning (DL): A subfield of ML using multi‑layer neural networks that can learn complex patterns from very large datasets.
• AI modelling: Developing algorithms (models) that can be trained to provide intelligent outputs from input data.

Types of AI Models

Two broad approaches to building models:

• Rule‑based models: Use explicitly defined rules and logic to make decisions.
• Learning‑based models: Learn patterns from labelled or unlabelled data using algorithms. Learning‑based models rely on training data and optimisation; they still use code and algorithms but adapt their behaviour based on data.

Decision Tree

• A decision tree builds classification or regression models in the form of a tree structure.
• It splits a dataset into smaller subsets while growing a tree of decision nodes and leaf nodes.
• The final model is a set of rules represented as paths from the root to leaves; each leaf gives a prediction.

Types of Learning

There are three primary learning paradigms in machine learning:

• Supervised learning
• Unsupervised learning
• Reinforcement learning

Supervised Learning

• The training dataset is labelled-each example includes input features and a correct output (label).
• A label is information used as a tag for data (for example, exam grades for students).
• Two common supervised tasks:
  • Classification: Predict a category (for example, spam vs not‑spam). Typically involves discrete labels and decision boundaries.
  • Regression: Predict a continuous value (for example, predicting salary after several years). Can be linear or non‑linear.

Unsupervised Learning

• Works with unlabelled data; the model finds structure, patterns or groupings in the data.
• Used to discover relationships, clusters and main features of the data without prior labels.
• Common unsupervised techniques:
  • Clustering: Grouping similar data points together (for example, customer segmentation).
  • Dimensionality reduction: Reducing the number of features while preserving important information. This helps visualise high‑dimensional data and speeds up models (common technique: Principal Component Analysis).

Reinforcement Learning

• Reinforcement learning concerns agents that learn to take actions in an environment to maximise cumulative reward.
• The agent explores actions, receives feedback (rewards or penalties), and updates its strategy to improve future rewards.
• Typical applications include game playing, robotics and control systems.