It works on unlabeled data.

It learns by trial and error based on a reward mechanism.

It requires labeled data for training.

It is based on a predefined set of rules.

Learning-Based Approach

Reinforcement Learning

Unsupervised Learning

Rule-Based Approach

Classification Model

Regression Model

Clustering Model

Association Model

They are based on a simple decision tree approach.

They require minimal data for training.

They are able to automatically extract data features.

They only work on numerical data.

Data Processing

Data Mining

Data Labeling

Data Extraction

Artificial Intelligence

Supervised Learning

Rule-Based Approach

All of the above

Input Layer

Hidden Layer

Output Layer

Supervised Learning

Reinforcement Learning

Classification

Clustering

Supervised Learning

Unsupervised Learning

Reinforcement Learning

Classification

Input Layer

Hidden Layer

Output Layer

All layers

Unsupervised Learning

Regression

Classification

Clustering

It requires vast amounts of data.

It is too complex to implement.

The learning is static and does not adapt to changes.

It cannot be used for chatbots.

Rows of a table

The output of a model

Columns of a table

The final prediction

Training data is unlabeled, while testing data is labeled.

Training data is used to teach the model, and testing data is used to evaluate its accuracy.

Training data contains only features, and testing data contains only labels.

There is no difference; they are interchangeable.

Regression model

Clustering model

Classification model

Association model

The weight

The coin's currency

The model's prediction

The training algorithm

Supervised Learning

Reinforcement Learning

Unsupervised Learning

Rule-Based Approach

It is based on static rules.

It learns from explicit programming.

It adapts to changes in data.

It does not require any data.

A type of neural network layer

A machine learning algorithm

A simplified model of how an AI makes a decision

A type of data

Classification

Regression

Clustering

Association

To process data and pass it to the hidden layers.

To perform calculations with weights and biases.

To present the final processed data to the user.

To acquire data and feed it to the network.

Classification

Clustering

Regression

Association

Supervised Learning

Unsupervised Learning

Reinforcement Learning

Rule-Based Learning

Unlabeled data

Only features

Only labels

Labeled data

It is too expensive to implement.

It fails to learn from its mistakes.

It requires continuous human intervention.

It can only handle a single type of data.

Machine Learning

Deep Learning

Rule-Based Approach

Supervised Learning

Input layer

Hidden layer

Output layer

Both (a) and (c)

An image

A spreadsheet

A text document

A database

Trial and error

Backpropagation

Forward propagation

Activation

Classification uses unlabeled data, while clustering uses labeled data.

Classification assigns objects to predefined classes, while clustering finds similarities and groups objects.

Classification is a supervised model, while clustering is a reinforcement model.

Clustering requires more data than classification.

To identify relationships in unlabeled data.

To predict a continuous value.

To make a series of decisions that maximize a reward.

To classify data into discrete categories.

Classification

Regression

Machine Learning

Reinforcement Learning

The words in the email

The sender's information

The classification of emails as either "spam" or "legitimate"

The email attachments

To train the model with new data

To check for errors in the code

To evaluate the accuracy of the trained model

To find hidden patterns

Predicting the price of a house

Predicting the number of days a patient will stay in a hospital

Predicting whether a patient will have a short or long hospital stay

Predicting a city's average temperature for the next month

Categorical data

Discrete data

Continuous data

Unlabeled data

Supervised Learning

Regression

Clustering

Reinforcement Learning

The human brain and nervous system

The human digestive system

A computer's hardware components

A mathematical formula

To create new features

To identify new patterns without guidance

To apply the knowledge to test data

To define the rules itself

Predictive-based

Reward-based

Learning-based

Both (b) and (c)

A column of a table

The name of the dataset

A tag that gives meaning to data

The algorithm used in the model

Machine Learning model

Rule-Based model

Deep Learning algorithm

Reinforcement Learning model

Regression model

Classification model

Association model

Clustering model

DL has more applications than AI.

All ML applications are also DL applications.

All AI applications are also ML applications.

DL has a very specific set of applications, a subset of ML, which is a subset of AI.

Supervised Learning

Unsupervised Learning

Regression

Classification

To acquire input data.

To give the final output.

To perform computations.

To act as a user interface.

Rule-based is static, while learning-based is adaptive.

Rule-based is more efficient for large datasets.

Learning-based is based on predefined rules.

There is no significant difference.

To define the final output.

To acquire input data.

To perform computations within the hidden layers.

To create a user interface.

To learn from a minimal amount of data.

To train itself to perform tasks with vast amounts of data.

To follow a predefined set of rules.

To predict discrete values only.

Both A and R are true, and R is the correct explanation of A.

Both A and R are true, but R is not the correct explanation of A.

A is true, but R is false.

A is false, but R is true.

Both A and R are true, and R is the correct explanation of A.

Both A and R are true, but R is not the correct explanation of A.

A is true, but R is false.

A is false, but R is true.

True

False

Both Assertion and Reason are true, and Reason is the correct explanation of Assertion.

Both Assertion and Reason are true, but Reason is not the correct explanation of Assertion.

Assertion is true, Reason is false.

Assertion is false, Reason is true.

Both Assertion and Reason are false.

Both Assertion and Reason are true, and Reason is the correct explanation of Assertion.

Both Assertion and Reason are true, but Reason is not the correct explanation of Assertion.

Assertion is true, Reason is false.

Assertion is false, Reason is true.

Both Assertion and Reason are false.

Both Assertion and Reason are true, and Reason is the correct explanation of Assertion.

Both Assertion and Reason are true, but Reason is not the correct explanation of Assertion.

Assertion is true, Reason is false.

Assertion is false, Reason is true.

Both Assertion and Reason are false.

Both Assertion and Reason are true, and Reason is the correct explanation of Assertion.

Both Assertion and Reason are true, but Reason is not the correct explanation of Assertion.

Assertion is true, Reason is false.

Assertion is false, Reason is true.

Both Assertion and Reason are false.

Both Assertion and Reason are true, and Reason is the correct explanation of Assertion.

Both Assertion and Reason are true, but Reason is not the correct explanation of Assertion.

Assertion is true, Reason is false.

Assertion is false, Reason is true.

Both Assertion and Reason are false.

Both Assertion and Reason are true, and Reason is the correct explanation of Assertion.

Both Assertion and Reason are true, but Reason is not the correct explanation of Assertion.

Assertion is true, Reason is false.

Assertion is false, Reason is true.

Both Assertion and Reason are false.

Both Assertion and Reason are true, and Reason is the correct explanation of Assertion.

Both Assertion and Reason are true, but Reason is not the correct explanation of Assertion.

Assertion is true, Reason is false.

Assertion is false, Reason is true.

Both Assertion and Reason are false.

Both Assertion and Reason are true, and Reason is the correct explanation of Assertion.

Both Assertion and Reason are true, but Reason is not the correct explanation of Assertion.

Assertion is true, Reason is false.

Assertion is false, Reason is true.

Both Assertion and Reason are false.

Both Assertion and Reason are true, and Reason is the correct explanation of Assertion.

Both Assertion and Reason are true, but Reason is not the correct explanation of Assertion.

Assertion is true, Reason is false.

Assertion is false, Reason is true.

Both Assertion and Reason are false.

Both Assertion and Reason are true, and Reason is the correct explanation of Assertion.

Both Assertion and Reason are true, but Reason is not the correct explanation of Assertion.

Assertion is true, Reason is false.

Assertion is false, Reason is true.

Both Assertion and Reason are false.

Both Assertion and Reason are true, and Reason is the correct explanation of Assertion.

Both Assertion and Reason are true, but Reason is not the correct explanation of Assertion.

Assertion is true, Reason is false.

Assertion is false, Reason is true.

Both Assertion and Reason are false.

Both Assertion and Reason are true, and Reason is the correct explanation of Assertion.

Both Assertion and Reason are true, but Reason is not the correct explanation of Assertion.

Assertion is true, Reason is false.

Assertion is false, Reason is true.

Both Assertion and Reason are false.

Both Assertion and Reason are true, and Reason is the correct explanation of Assertion.

Both Assertion and Reason are true, but Reason is not the correct explanation of Assertion.

Assertion is true, Reason is false.

Assertion is false, Reason is true.

Both Assertion and Reason are false.

Both Assertion and Reason are true, and Reason is the correct explanation of Assertion.

Both Assertion and Reason are true, but Reason is not the correct explanation of Assertion.

Assertion is true, Reason is false.

Assertion is false, Reason is true.

Both Assertion and Reason are false.

Both (A) and (R) are true, and (R) correctly explains (A).

Both (A) and (R) are true, but (R) does not explain (A).

(A) is true, but (R) is false.

(A) is false, but (R) is true.

Both (A) and (R) are false.

Both (A) and (R) are true, and (R) correctly explains (A).

Both (A) and (R) are true, but (R) does not explain (A).

(A) is true, but (R) is false.

(A) is false, but (R) is true.

Both (A) and (R) are false.

Both (A) and (R) are true, and (R) correctly explains (A).

Both (A) and (R) are true, but (R) does not explain (A).

(A) is true, but (R) is false.

(A) is false, but (R) is true.

Both (A) and (R) are false.

A Rule-Based Approach is where a developer explicitly defines rules and instructions; a Learning-Based Approach is where the machine learns patterns on its own and adapts to changes.

A Rule-Based Approach relies only on labeled data; a Learning-Based Approach relies only on unlabeled data.

A Rule-Based Approach improves its accuracy over time; a Learning-Based Approach remains static and does not adapt.

A Rule-Based Approach automatically discovers rules; a Learning-Based Approach strictly follows fixed instructions.

Supervised Learning, Unsupervised Learning, Reinforcement Learning

Classification, Regression, Clustering

Deep Learning, Neural Networks, Decision Trees

Training Data, Testing Data, Validation Data

Labeled data has a tag or label attached, while unlabeled data has no tags attached.

Labeled data is always numerical, while unlabeled data is always text.

Labeled data is collected from sensors, while unlabeled data is collected from surveys.

Labeled data is used only for testing, while unlabeled data is used only for training.

Features are the columns describing data, and a label is the target variable to be predicted; for example, predicting fruit type based on color where color is the feature and fruit name is the label.

Features are the target outcomes, and labels are the input variables; for example, fruit name is the feature and color is the label.

Features and labels are identical; both represent the target to be predicted in a dataset.

Features are tags attached to records, while labels are metadata about the dataset structure.

A training dataset is used to fit the model, and a testing dataset is used to evaluate the model on unseen data.

A training dataset is used only for feature selection, and a testing dataset is used to tune hyperparameters.

A training dataset evaluates model performance, and a testing dataset is used to train the model.

A training dataset is used for storing labels, and a testing dataset is used for storing features.

A system of machine learning algorithms modeled after the human brain and nervous system that automatically extracts data features without programmer input

A pre-processing tool that cleans datasets before training other models

A rule-based, static program that does not learn from new data

A visualization technique used only to display neural connections

Classification models and Regression models

Clustering and Dimensionality Reduction

Reinforcement Learning and Unsupervised Learning

Association Rule Mining and Anomaly Detection

Predicting whether an email is spam or not spam; Predicting the price of a house based on its features

Predicting the price of a house; Grouping customers by purchasing behavior

Detecting anomalies; Predicting whether an email is spam or not spam

Clustering emails by topic; Estimating the probability of rainfall as a label

A child learns to swim on his own without any supervision, discovering patterns independently

A teacher provides graded answers for every homework problem

A robot follows a fixed set of hand-written rules to perform a task

A model trained only with labeled examples and explicit targets

A simplified model of how an AI makes a decision that takes multiple inputs, weights them, adds a bias, and compares the result to a threshold to produce an output

A memory unit that stores training data for later retrieval

A clustering algorithm used to group unlabeled data

A visualization tool for displaying neural network architectures

The input layer acquires data and feeds it into the network; the output layer presents the final processed data to the user; neither layer performs processing

Both input and output layers perform most of the data processing

The input layer presents final results while the output layer collects raw data

Only hidden layers feed data into the network; input and output layers are optional

It takes factors as inputs, assigns weights based on importance, sums them with a bias, and compares the result to a threshold

It selects outputs randomly to avoid bias

It uses a single hard-coded rule regardless of input

It treats all inputs with equal weight and ignores bias

It is adaptive and can modify itself and learn from new data, whereas a Rule-Based model is static and does not learn from changes

It requires no data to function, unlike Rule-Based models

It guarantees perfect accuracy on all tasks

It is deterministic and cannot change behavior after deployment

To identify relationships, patterns, and trends from an unlabeled dataset by finding similarities and differences

To predict labeled targets provided during training

To execute predefined rules without learning

To control robots using reward signals

A piece of information or tag attached to a data point to give it meaning—the known target the model is trained to predict

A learned weight used inside the model to score features

A raw feature measured from the data source

An error metric used to evaluate model performance

A specific learning approach and subset of Machine Learning that enables software to train itself to perform tasks with vast amounts of data by using multiple machine learning algorithms working together

A rule-based programming method that avoids learning and relies on hand-written logic

A technique limited to small datasets and a single algorithm

A data visualization approach that represents neural network outputs only

A system that assigns emails to predefined folders like "spam" or "inbox"

A supermarket analysis that finds customers who buy bread are also likely to buy butter

A model that predicts tomorrow’s temperature from historical readings

A program that groups handwritten digits after being told the correct label for each

Classification assigns data to predefined classes; Clustering groups similar data without predefined classes

Classification groups data without labels; Clustering requires labeled data

Classification is always unsupervised; Clustering is always supervised

Classification uses continuous outputs; Clustering predicts numeric values

They store the training dataset for reuse

They control the influence of inputs and are adjusted during learning to reduce error and produce accurate outputs

They label the input data for supervised training

They prevent the network from overfitting by deleting layers

Regression uses categorical data; Classification uses continuous data

Regression uses continuous numeric data; Classification uses discrete categorical data

Regression and Classification both require labeled categorical data

Regression uses images; Classification uses text

AI is a subset of ML, and DL is broader than AI

AI is the broadest field; ML is a subset of AI; DL is a subset of ML forming a narrowing funnel

ML is the broadest field; AI is a subset of ML; DL is unrelated

DL is broader than ML and AI and does not use neural networks

Learning from unlabeled data to discover hidden groups, such as clustering customers by behavior

Learning from labeled data that pairs features with labels, such as predicting a coin’s currency from its weight with the currency as the label

Learning by trial and error without any data

Learning from rules written by experts with no training data

A Rule-Based Approach dynamically adapts; a Learning-Based Approach is static

Both approaches are static and follow fixed rules

A Learning-Based Approach is dynamic and adapts by modifying algorithms with data, while a Rule-Based Approach follows predefined rules and does not adapt

A Learning-Based Approach does not use data; a Rule-Based Approach learns from data

Input performs all computations; hidden collects data; output stores training data

Input acquires data; hidden layers perform computations and processing; output produces the final results

Input is the label; hidden layers are unused; output selects features

Input generates predictions; hidden layers handle data storage; output trains the network

Classification and Regression; classification predicts discrete categories while regression predicts continuous values

Clustering and Association; clustering groups unlabeled data while association finds relationships in a database

Reinforcement and Unsupervised; reinforcement uses rewards and punishments while unsupervised uses labeled data

Anomaly Detection and Classification; anomaly flags rare events while classification predicts continuous outputs

It learns through trial-and-error, receiving negative feedback for incorrect choices and positive feedback for correct ones, improving decisions based on rewards and punishments

It groups similar data points into clusters using similarities without any feedback

It calculates the error once and sets fixed rules that never change

It relies on predefined labels to classify inputs without any exploration

It analyzes unlabeled data to identify relationships, patterns, and trends, often forming clusters based on similarities without prior guidance

It uses labeled data to classify inputs into predefined categories

It follows fixed rules to generate outputs without changing weights

It depends on rewards and punishments to learn optimal actions

Data passes from the input layer through hidden layers where weights and biases are adjusted via repeated trial-and-error and error feedback until the network learns the correct output

Weights are set once at initialization and never updated

Outputs are determined solely by predefined rules without hidden layers

The model guesses randomly until it finds the correct answer, without using error calculations

Clustering groups similar unlabeled data points (e.g., OTT platform recommending songs with similar tempo), and Association finds relationships in data (e.g., supermarket linking bread buyers to butter purchases)

Classification and Regression predict categories and continuous values with labeled training data

Reinforcement and Supervised learning optimize actions using rewards and labeled targets

Anomaly Detection and Clustering both flag outliers as anomalies in labeled datasets

They learn from new data and past mistakes, account for exceptions, and adapt over time rather than relying on fixed rules

They depend entirely on static, rule-based logic

They only improve when labels are removed from the dataset

They avoid considering errors to prevent overfitting

It learns normal patterns in a dataset and flags data points that deviate significantly from those patterns as anomalies

It classifies inputs into predefined categories using labeled examples

It clusters data points purely based on similarity without defining normal behavior

It uses rewards and punishments to choose optimal actions

Regression predicts continuous values within a range, while classification assigns data to discrete categories.

Regression assigns data to classes like "spam/not spam," while classification predicts a temperature on a continuous scale.

Both regression and classification predict continuous values with no categorical outputs.

Regression and classification are both unsupervised methods that do not use labeled data.

Predicting whether a customer is eligible for a bank loan; features include credit score, income, employment history; label is "eligible" or "not eligible."

Clustering users based on browsing patterns without labels; features and labels are unknown.

Detecting anomalies in transactions without a predefined "fraud" label; the model learns patterns without labels.

Predicting weather types using unlabeled temperature readings; the label is the temperature itself.

A learning-based filter is trained on labeled emails ("spam"/"legitimate"), learns content patterns, and adapts to new examples, whereas a rule-based filter relies on predefined rules and fails to catch novel spam.

A rule-based filter is more effective because fixed rules always generalize better than learned patterns.

A learning-based filter does not require labeled emails and therefore cannot adapt its performance over time.

A learning-based filter only applies static rules and is less accurate than a rule-based filter.

A social media platform identifying a friend in a tagged photo album is supervised learning.

An OTT platform recommending content based on unlabeled watch history is unsupervised learning.

A bank analyzing transactions for suspicious activity without a predefined "fraud" label is unsupervised learning.

Detecting spam solely by fixed rules is supervised learning.

A training data set is a collection of labeled examples used for a model to learn; it is similar to a teacher-student relationship because, like instructional examples before testing, labeled data teaches the model before evaluation.

A training data set is unlabeled and is only used after the model is deployed.

A training data set contains a single example; students are graded without prior practice, mirroring model training.

It is similar to a teacher-student relationship because students generate labels for the teacher during instruction.

Features are descriptive attributes or columns (e.g., color, size, price); a label is the target variable the model is trained to predict.

A feature is the final decision output, while a label is any column in the dataset.

Features are tags or target variables, whereas labels are descriptive attributes.

There is no difference; features and labels are interchangeable terms.

Weights represent the importance of each input (higher weight increases influence), and bias controls the decision threshold or cautiousness (higher bias makes decisions more conservative).

Weights and bias are random values and do not affect the decision outcome.

Bias always equals the sum of weights, and a higher bias always increases acceptance of inputs.

Weights represent labels, and bias is an input feature.

It is the task of identifying and labeling objects within an image.

It uses algorithms to analyze what is in a picture and categorize those things.

An example given is a model that correctly predicts a bird when pixels of a bird image are input.

It is primarily about discovering clusters in unlabeled tabular data.

A CNN is a type of Deep Learning algorithm.

CNNs are specifically designed to process image-based data.

CNNs assign importance (via learnable weights and biases) to different aspects of an image to differentiate features.

There is no difference between CNNs and ANNs; they are identical in purpose and design.

CNNs are highly effective for tasks like object identification and image classification.

Supervised Learning requires a labeled dataset and aims to predict an output based on input features.

Unsupervised Learning works on unlabeled data to discover hidden patterns or groups.

Reinforcement Learning involves trial-and-error interactions with an environment, using rewards for good actions and penalties for bad ones.

A practical example of Reinforcement Learning is a robot receiving positive rewards for moving toward an exit and negative rewards for hitting a wall.

Unsupervised Learning provides explicit labels during training to guide the model.

An ANN consists of three main types of layers: an input layer, one or more hidden layers, and an output layer.

The input layer acquires raw data and passes it to the next layer without performing computations.

Hidden layers perform the actual computations, using weights and biases to adjust the output.

The output layer provides the final output to the user and does not perform computation like the hidden layers.

Training involves a continuous loop of trial and error where weights are adjusted based on error to improve accuracy.

A developer explicitly programs decision rules (e.g., a decision tree with fixed instructions); it is static because it cannot learn from new data; a key limitation is poor adaptability to unforeseen scenarios.

A model discovers rules from data without human guidance; it is static because it only uses historical data; a key limitation is excessive computational cost.

A developer programs only high-level goals and the system learns rules autonomously; it is static because training takes long; a key limitation is overfitting to training data.

A model uses probabilistic learning to update rules after each interaction; it is static because updates happen infrequently; a key limitation is needing large datasets.

Classification predicts a discrete or categorical output; a bank default prediction is suitable because the output is yes/no.

Regression predicts a continuous numerical value; house price prediction is suitable because the output varies within a range.

Classification predicts continuous values; it is suitable for pricing houses because the target is a number.

Regression assigns customers to one of two categories; it is suitable for default prediction because the task is categorical.

Unsupervised learning operates on unlabeled data to discover hidden patterns without guidance.

Clustering groups data into clusters based on similarities and is used to organize users/items with shared characteristics.

Association finds interesting relationships or "association rules" between variables, such as products frequently purchased together.

Supervised learning requires labeled outputs and focuses on predicting predefined targets from inputs.

Each input (e.g., "Do I have a jacket?", "Is it sunny now?", "Do I have an umbrella?", "What is the weather forecast?") is assigned a weight; the perceptron adds the weighted inputs and a bias, compares the total to a threshold, and outputs yes/no accordingly.

The perceptron averages all inputs, ignores any bias, and outputs yes if the average exceeds 0.5.

The perceptron selects the single most important input and uses a majority vote across repeated checks to output yes/no.

Inputs are multiplied together without weights; if the product is nonzero, the perceptron outputs yes, otherwise no.

Data

Features

Labels

Training Dataset

Testing Dataset

Data

Features

Labels

Training Dataset

Testing Dataset

Data

Features

Labels

Training Dataset

Testing Dataset

Data

Features

Labels

Training Dataset

Testing Dataset

Data

Features

Labels

Training Dataset

Testing Dataset

An approach where an agent learns by trial and error to maximize cumulative reward through feedback for desired and undesired actions

A method that relies exclusively on large labeled datasets to fit a mapping from inputs to outputs

A technique that clusters data points based on similarity without any feedback signals

A supervised approach that minimizes classification error using fixed training labels

It can adapt by exploring the environment and learning strategies from feedback without needing extensive pre-existing labeled datasets

It requires a comprehensive labeled dataset before any learning can begin

It is only effective when the environment is static and fully known in advance

It replaces trial-and-error with direct instruction from a fixed teacher

Classification, because the output is a discrete category (short stay or long stay)

Regression, because the output is a continuous number of days

Classification, because the output is a continuous number of days

Regression, because the output is a pair of categories

Regression, because the output is a continuous value (number of days)

Classification, because the output is a discrete category (short stay or long stay)

Classification, because the output is a probability of a class

Regression, because the output is a categorical label

Supervised learning involves guidance from labeled data during training, while unsupervised learning proceeds without such guidance

Both supervised and unsupervised learning require labeled data and a teacher

Unsupervised learning uses labels provided by a teacher, while supervised learning discovers patterns without labels

Supervised learning and unsupervised learning both avoid feedback or guidance entirely

Predict a known outcome from labeled examples

Discover clusters in an unlabeled dataset

Generate new synthetic data without guidance

Randomly assign labels to unseen data

A teacher guiding a student with worked examples

A child learning to swim alone without supervision

An explorer wandering without a map

A machine guessing patterns with no feedback

Labeled dataset with target outcomes

Partially labeled dataset with some targets

Unlabeled dataset with no provided answers

Balanced dataset with equal class sizes

Predict known labels on unseen data

Memorize training examples for perfect recall

Find inherent structure like clustering similar points

Reduce dataset size through compression

It serves as a teacher by providing solved examples

It is ignored so the model discovers patterns alone

It is used only to generate synthetic samples

It is removed to avoid bias during training