Machine Learning Algorithms

Machine learning in a nutshell

• Subset of AI, enables systems to learn and improve without explicit programming

• Data is critical to ML

• More data = more accurate prediction

Key features

• Interpretation, analysis, and processing of data to make it applicable

• Learns from data; enhances over time

• Facilitates automation and prediction

• Prevailing approach in contemporary AI

Applications of KNN

• Image recognition

  • Categorize photographs depending attributes(color, pixels, etc.)

  • Compare attributes with the training set

• Spam detection

  • Comparison of new emails to a database containing spam and non-spam

Applications of linear regression

• Prediction of product demand

• Sales forecast

• Prediction of revenue through ads

Positive correlation

• Linear relationship b/w 2 variables

  • One goes up, the second does too

• Demand increases, so does supply

Negative correlation

• Relationship b/w 2 variables

  • One decreases, then the other increases, and vice versa

• More calories burned = less weight

No correlation

• No relationship b/w variables

• Amount of candy consumed and intelligence

Non-linear correlation

• All points of a scatter plot tend to lie near a smooth curve

Correlation

• Determines the strength or degree of relationship b/w 2 variables

• Represented by a single value

Regression

• Determines how 1 variable affects another

• Represented by a regression line

• Predicts continuous values

Classification

• Label a set of data into different classes/groups

  • each class can be labelled

• Goal: determine which class new data will belong to

Limitations of KNN

• Requires careful tuning

  • Appropriate k selection has a significant impact on accuracy

• Imbalanced datasets

  • Biased towards dominant class

• Highly computational evaluation stage

  • Calculates distance for every training

• Large memory required

  • Stores every piece of training data

Centroid

• Imaginary/ real location denoting centre of the cluster

Causation

• An event is the direct result of another

• Cause and effect

• Sleep deprivation leads to brain fog

Outlier

• Data points significantly different from other observations 

• Measurement irregularity or experimental error

Centroid-based clustering

• Arranges data into non-hierarchical clusters

• K-means clustering

• Efficient; easily affected by initial conditions or outliers

Density-based clustering

• High density areas into clusters

• Arbitrary-shaped distributions occur

• Data points, separating regions, low density considered outliers

  • Not assigned clusters

Hierarchical clustering

• Builds tree of cluster

• Tiered series of nested clusters

• Each cluster diff. from every other cluster

  • Objects within each cluster similar to each other

Binary Classification

• Classification tasks with two class labels.

• Email spam detection

Multi-Class Classification

• more than two class labels.

• Face classification

Multi-Label Classification

• Each example may belong to multiple class labels.

• • Photo classification

Imbalanced Classification

• unequally distributed class labels

  •  a majority and minority class.

• Medical diagnostic tests