3. FC | Machine Learning Foundations

Definition of Machine Learning: A subset of artificial intelligence (AI) that enables computer systems to learn from data and make predictions without explicit programming.
Types of Machine Learning:
- Supervised Learning: Involves the use of labeled data to train models.
- Example Algorithms: Linear Regression and Logistic Regression.
- Unsupervised Learning: Learns from unlabeled data to identify underlying patterns.
- Example Technique: Clustering.
- Reinforcement Learning: Learns through interaction with the environment using rewards and penalties.
Key Concepts:
- Training and Inference: Training involves teaching a model using data, whereas inference is using the trained model to make predictions.
- Loss Function: Measures the difference between predicted and actual values to optimize model performance.

What distinguishes regression from classification in supervised learning?
How do machine learning models learn relationships between input features and output labels?
Why is standardization crucial in machine learning preprocessing?
How does clustering function without labeled data?
What is the role of rewards and penalties in reinforcement learning?

Machine learning empowers computers to learn from data incorporating algorithms that are intelligent through examples.
Supervised learning uses labeled datasets:
- Regression: Predicts continuous outputs (e.g., linear regression for predicting values).
- Classification: Assigns outputs to categories (e.g., logistic regression for binary classification).
Linear Regression: A method for predicting continuous outcomes by fitting a line through data points.
- Example Equation: $f(x) = w imes x + b$ where:
- $w$ = slope
- $b$ = y-intercept
Logistic Regression: A classification method that uses a sigmoid function to predict probabilities.
- Sigmoid function ranges from 0 to 1 for interpreting probability outcomes.
- Example: For a student pass/fail prediction, if the output exceeds a threshold (commonly 0.5), then classified as “pass”.

Concept: Supervised learning approach using labeled data to map input features to output labels.
Application: House price estimation:
- Independent feature: Size of the house in square feet.
- Dependent feature: Price in dollars.
Fitting the Model:
- Iteratively adjusts parameters (slope and intercept) to minimize the difference between actual and predicted values (loss).
- Loss Calculation: Squared difference between predicted and actual values.

Classification: Supervised learning method for categorical outputs.
- Binary Classification: Two classes (e.g., spam vs. not spam).
- Multi-class Classification: More than two categories (e.g., positive, negative, and neutral sentiments).
Logistic Regression:
- A classification algorithm predicting binary outcomes using a sigmoid function.
- Example: Student pass/fail prediction based on study hours using sigmoid function for probability.
Iris Dataset: Demonstration of multi-class classification involving three types of iris flowers based on different attributes.

Machine Learning Workflow Phases:
1. Data Loading
2. Data Preprocessing
3. Model Training
4. Model Evaluation
5. Making Predictions
Data Preprocessing: Key actions like handling missing values and normalizing data.
- Standardization: Adjusting data to a mean of 0 and standard deviation of 1 to prevent larger features from skewing results.
Train-Test Split: Validation method splitting data for training and testing.
- The model is assessed on the testing set to ensure generalization to unseen data.
- Evaluation Metric: Accuracy score measuring the correctness of predictions.

Unsupervised Learning: Operates without labeled outputs, discovering data patterns through clustering.
Clustering: Groups similar items, identifying outliers that do not fit well with others, based on their similarities.
Similarity Metrics: Various metrics calculate closeness of data points, such as:
- Euclidean Distance
- Manhattan Distance
- Cosine Similarity
- Jaccard Similarity
Real-world Applications:
- Market Segmentation for targeted advertising.
- Outlier analysis in fraud detection.
- Recommendation systems (e.g., Netflix clustering users).

Reinforcement Learning: Agent learns through feedback from an environment (rewards/penalties) without labeled data.
Key Terms:
- Agent: Learner making decisions.
- Environment: External system interaction.
- State: Current situation.
- Actions: Choices available to the agent.
- Policy: Strategy managing actions based on states.
Goal: Discovering an optimal policy for maximizing long-term cumulative rewards.
Algorithms: Q-learning and Deep Q-learning aid the agent's decision-making process.
Practical Applications: Examples include:
- Autonomous vehicles making real-time driving decisions.
- Smart home devices adapting to user preferences.
- Robotic performance optimization in manufacturing.
Case Study: A robotic arm learns to pick and place items efficiently via rewards for successes and penalties for failures.

Anaconda: An open-source platform for Python and R, ideal for data science projects with package management.
Jupyter Notebook: Interactive environment for live code and visualizations.
Iris Dataset (iris.csv): Standard dataset used in machine learning containing features of 150 iris flower instances.
Python Libraries:
- Pandas: For data manipulation.
- NumPy: For numerical computations.
- Scikit-learn: Provides machine learning algorithms, including Logistic Regression, traintestsplit, StandardScaler, and accuracy_score.
Important Note: AI systems may produce inaccuracies, and assessments for accuracy are crucial.