Machine Learning Basics

Types of ML: -

Supervised Learning

• You teach the model using labeled data (inputs + correct outputs)
• Goal : Predict outputs for new data
• Example: - Input = [size of house, number of rooms, location] - Output = price of house

Algorithms: -

• Linear Regression : Predicition
• Logistic Regression : Classification
• Decision Trees : Decision-Making
• Random Forests, SVM

Unsupervised Learning

• The model finds patterns without labels
• Goal : Understand structure in data
• Example : Customer segmentation = Group simliar customers for marketing

Algorithms: -

• K-Means : Clustering
• PCA : Dimensionality reduction

Reinforcement Learning (RL)

• A model (agent) learns by trial-and-error, receiving rewards.
• Goal : Maximize cumulative reward over time.
• Example : A robot learns to walk

Key terms: -

• Agent : learner/decision maker
• Environment : World interacted by the Agent
• Reward : Feedback for actions
• Policy : Strategy followed by the Agent

Deep Learning

Key Terms: -

Neural Networks (NNs)

• Layers of neurons: input -> hidden -> output
• Each neuron does a calculation, applies an activation function
• Example : Predict if an image is of a cat or dog

Convolutional Neural Networks (CNNs)

• Specifically of images
• Detect features like edges, textures, shapes using convolution layers
• Typical Structure : Convolution -> Pooling -> Fully Connected -> Output

Recurrent Neural Networks (RNNs)

• Specifically for sequential data like text, time series
• Maintains memory of previous inputs
• Variants: LSTM, GRU (solve long-term memory problems)

Transformers

• Modern architecture of NLP
• Uses attention mechanism -> understands context in whole sentence
• Example : GPT, BERT

Mathematics used in ML

ML uses math for data representation, predictions and optimization

Linear Algebra

• Vectors, matrices, dot products
• Used for representing data and weights in neural networks

Probability & Statistics

• Mean, variance, probability distributions
• Used in making predictions and understanding uncertainty

Calculus Basics

• Derivative -> used in gradient descent to minimize errors
• Partial derivatives -> used to update weights in NNs

Data Handling

Before feeding data to a model, it must be cleaned and prepared

Preprocessing

• Scale/normalize features
• Encode categorical variables (eg Red = 1, Blue = 2)

Cleaning

• Remove duplicates, correct errors
• Handle missing data (fill with mean/median or drop rows)

Feature Engineering

• Create new meaningful features from existing data
• Example -> Extract day/month/year from a date, compute BMI from height & weight.