CI

Computational Intelligence (CI)

A branch of Artificial Intelligence that focuses on adaptive, learning-based, and nature-inspired techniques to solve complex problems.

Artificial Intelligence (AI)

The broader field of making machines act intelligently, such as reasoning, learning, planning, recognizing patterns, and making decisions.

Machine Learning (ML)

A branch of AI where computers learn patterns from data and use them to make predictions or decisions.

Relationship between AI, ML, and CI

AI is the broadest field. ML focuses on learning from data. CI includes adaptive techniques such as neural networks, fuzzy logic, genetic algorithms, and swarm intelligence.

Is CI the same as ML?

No. CI and ML overlap, especially in neural networks, but CI also includes fuzzy logic, genetic algorithms, and swarm intelligence.

Example of AI

A chatbot, self-driving car, face recognition system, or expert system.

Example of ML

A model that learns from Iris flower data to classify the flower species.

Example of CI

An artificial neural network, fuzzy logic controller, genetic algorithm, or particle swarm optimization system.

Application of CI in healthcare

CI can be used to detect diseases, classify medical images, or predict patient risk.

Application of CI in agriculture

CI can be used for fruit grading, plant disease detection, crop prediction, or smart irrigation.

Application of CI in transportation

CI can be used in traffic prediction, self-driving systems, and route optimization.

Application of CI in energy

CI can be used to predict electricity consumption and optimize energy usage.

Classification

A machine learning task where the output is a category or class.

Regression

A machine learning task where the output is a continuous numerical value.

Classification example

Predicting whether an Iris flower is Setosa, Versicolor, or Virginica.

Regression example

Predicting daily electricity consumption in kWh.

Main difference between classification and regression

Classification predicts a class/category, while regression predicts a numerical value.

Supervised learning

A learning method where the model learns from labelled data with known correct answers.

Unsupervised learning

A learning method where the model finds patterns in unlabelled data.

Labelled data

Data that includes both input values and the correct output.

Artificial Neural Network (ANN)

A machine learning model inspired by biological neurons, used to learn patterns from data.

Neural Network (NN)

Another name commonly used for Artificial Neural Network.

ANN as a model

ANN is a model or technique used in machine learning and computational intelligence.

Backpropagation

A learning algorithm used to train ANN by calculating errors and updating weights.

Gradient Descent

An optimization method used to adjust weights to reduce error.

Neuron

A processing unit in ANN that receives inputs, applies weights, calculates a net value, and produces an output.

Input node

A node that receives input data or features.

Hidden neuron

A neuron in the hidden layer that processes information between input and output layers.

Output neuron

A neuron that produces the final prediction of the network.

Weight

A value that controls the strength of connection between neurons.

Bias

An additional value added to the weighted sum to shift the activation function.

Input layer

The first layer of ANN that receives the input features.

Hidden layer

The middle layer of ANN that learns patterns from the input data.

Output layer

The final layer of ANN that produces the prediction or classification.

ANN architecture

The structure of ANN, usually consisting of input layer, hidden layer, and output layer.

Example ANN architecture for Iris classification

4 input nodes, hidden layer, and 3 output nodes.

Why Iris ANN has 4 input nodes

Because the Iris dataset uses sepal length, sepal width, petal length, and petal width.

Why Iris ANN has 3 output nodes

Because there are three Iris species: Setosa, Versicolor, and Virginica.

Example ANN architecture for electricity prediction

4 input nodes, hidden layer, and 1 output node.

Why electricity prediction ANN has 1 output node

Because the model predicts one numerical value: daily electricity consumption.

Activation function

A function that transforms the net input of a neuron into an output.

Sigmoid activation function

A function that converts input into a value between 0 and 1.

Sigmoid formula

f(x) = 1 / (1 + e^-x)

Step function

An activation function that outputs 1 if the input reaches the threshold, otherwise outputs 0.

Threshold

A boundary value used by a perceptron to decide whether the output is 0 or 1.

Net input

The weighted sum of inputs before applying the activation function.

Net input formula

net = x1w1 + x2w2 + x3w3 + ...

Learning rate

A value that controls how much the weights change during training.

Small learning rate effect

The model learns slowly because weights change only slightly.

Large learning rate effect

The model may overshoot the best solution and become unstable.

Epoch

One complete pass through the training dataset.

Target output

The correct or desired output used during training.

Actual output

The output produced by the model.

Error

The difference between target output and actual output.

Simple error formula

Error = Target - Output

Loss function

A function used to measure how wrong the model prediction is.

Training data

Data used to train the model.

Testing data

Data used to evaluate the model after training.

Validation data

Data used to tune model settings before final testing.

Overfitting

When the model performs well on training data but poorly on new data.

Underfitting

When the model is too simple and fails to learn the pattern in the data.

Generalization

The ability of a model to perform well on unseen data.

Normalization

Scaling input values to a similar range so the model can learn better.

Why normalize ANN input

To prevent features with larger values from dominating the learning process.

Hyperparameter tuning

The process of adjusting settings such as learning rate, hidden neurons, epochs, and activation function.

ANN process step 1

Prepare and clean the dataset.

ANN process step 2

Divide the dataset into training and testing sets.

ANN process step 3

Choose ANN architecture such as number of input, hidden, and output nodes.

ANN process step 4

Initialize weights, usually with small random values.

ANN process step 5

Perform forward propagation to calculate output.

ANN process step 6

Calculate error between target and actual output.

ANN process step 7

Use backpropagation to update weights.

ANN process step 8

Repeat training for several epochs until error is reduced.

Forward propagation

The process of passing input through the network to calculate the output.

Backpropagation

The process of sending error backward through the network to update weights.

K-Fold Cross Validation

A model evaluation method where data is split into K parts, and each part is used once as testing data.

Purpose of K-Fold Cross Validation

To evaluate model performance more reliably by testing the model on different data splits.

Example of 5-Fold Cross Validation

The dataset is split into 5 parts; the model trains on 4 parts and tests on 1 part, repeated 5 times.

Advantage of K-Fold Cross Validation

It reduces bias from using only one train-test split.

Confusion Matrix

A table that compares actual classes with predicted classes.

True Positive (TP)

The model predicts positive and the actual class is positive.

True Negative (TN)

The model predicts negative and the actual class is negative.

False Positive (FP)

The model predicts positive but the actual class is negative.

False Negative (FN)

The model predicts negative but the actual class is positive.

Accuracy

The proportion of total predictions that are correct.

Accuracy formula

Accuracy = (TP + TN) / (TP + TN + FP + FN)

Precision

The proportion of predicted positive cases that are actually positive.

Precision formula

Precision = TP / (TP + FP)

Recall

The proportion of actual positive cases that are correctly detected.

Recall formula

Recall = TP / (TP + FN)

F1 Score

A performance measure that balances precision and recall.

F1 Score formula

F1 = 2 × (Precision × Recall) / (Precision + Recall)

Precision simple meaning

When the model says positive, how often is it correct?

Recall simple meaning

Out of all actual positives, how many did the model find?

Accuracy simple meaning

Out of all predictions, how many were correct?

When accuracy can be misleading

When the dataset is imbalanced and one class is much larger than another.