Machine Learning Concepts Quiz

Machine Learning Overview

• Machine Learning is a field of artificial intelligence enabling systems to learn and improve from experience without explicit programming.
• It's based on computers learning from data, identifying patterns, and making judgments with minimal human intervention.
• Machines learn and improve their behaviour and decisions via automated learning processes.
• Data quality is essential for the accuracy of machine learning models.
• Algorithm selection depends on the nature of data and desired activity.

Human Learning vs. Machine Learning

Machine Learning Terminology

• Model: A mathematical representation of a real-world process learned from data.
• Feature: A measurable property of the data.
• Feature Vector: A set of multiple numeric features used as input to the model.
• Training: The process of fitting a model to data to learn patterns.
• Prediction: Using the trained model to predict outputs for new inputs.
• Target/Label: The value the model predicts or aims to understand.
• Overfitting: When a model learns the training data too well, including noise and inaccurate data points, decreasing performance on new data.
• Underfitting: When a model doesn't learn the underlying trend in the data.

Machine Learning Workflow

• Data Collection: Gathering relevant data from various sources.
• Data Preparation: Cleaning, transforming, and preparing the raw data for modeling.
• Choosing Learning Algorithm: Selecting the most suitable algorithm for the task based on data type and problem.
• Training Model: Building the model using training data.
• Evaluating Model: Assessing the model's performance on unseen data.
• Predictions: Using the model to make predictions on new, unseen data.

Artificial Intelligence vs. Machine Learning

Types of Machine Learning

• Supervised Learning: Uses labeled data to train a model to predict targets based on inputs. (e.g., classification and regression)
• Unsupervised Learning: Learns patterns and relationships from unlabeled data, with no pre-defined outputs. (e.g., clustering)
• Reinforcement Learning: An agent learns to make decisions in an environment by receiving rewards or penalties for its actions in that environment.

Tools and Technology for Machine Learning

• Programming Languages: Python (widely used due to its extensive toolkits), R.
• Libraries and Frameworks: Scikit-learn, TensorFlow, PyTorch, Keras.
• Data Processing and Analysis: NumPy, Pandas.
• Visualization Tools: Matplotlib, Seaborn, TensorBoard.

Applications of Machine Learning

• Facial Recognition: Security, crime investigations, etc.
• Speech Recognition: Voice-activated assistants, converting speech to text.
• Financial Services: Fraud detection, credit scoring, trading decisions.
• Healthcare: Disease diagnosis, drug discovery.
• Traffic Predictions: Improving travel times/route optimization.

Preparing the Model, Modeling, and Evaluation

• Selecting a Model: Choosing the best model based on data, target, and random factors.
• Training a model: Input data, training processes, to result in a model.
• Cross-validation: Method to measure model performance. (e.g., Holdout method, K-fold cross-validation)
• Model representation: Interpreting model structure.

Probability and Statistics

• Probability: Measures the likelihood of an event occurring.
• Statistics: Analyzes data to derive insights about how frequently things happen and relationships between factors.
• Descriptive Statistics: Organize, summarize, and describe data, using measures like mean, median, mode, variance, and standard deviation.
• Inferential Statistics: Use sample data to make inferences and draw conclusions about a larger population including hypothesis testing for decision-making.
• Random Variables: Represents outcomes of random trials.
• Probability Distributions (Discrete/ Continuous): Shows the probabilities associated with possible values of random variables.
• Central Limit Theorem: Explains how the distribution of sample means approximates a normal distribution for large samples.
• Monte Carlo Simulation: Technique using random sampling to model uncertain outcomes.

Classification Algorithms

• Supervised Learning: Learning with labeled training data.
• Classification: Categorizing data into predefined classes based on input attributes.
• Linear Models: Models (e.g., Logistic Regression) with a linear decision boundary.
• Non-linear Models: Models (e.g., Support Vector Machines) with non-linear decision boundaries.
• K-Nearest Neighbors (KNN): Classifies new data points based on the categories of the 'k' nearest neighbors in the training data.
• Naive Bayes: Probabilistic classifier based on Bayes' Theorem, making independent prediction probabilities.

Regression Algorithms

• Supervised learning: Predicting continuous values.
• Regression Models: Modeling the relationship between input and output variables.
• Linear Regression: Simple linear relationship between variables.
• Polynomial Regression: Curvilinear relationship between variables.

Decision Trees

• Classification and Regression: Decision tree models can solve classification and regression problems.
• Decision Nodes: Internal nodes in the tree that ask questions based on features.
• Leaf Nodes: Terminal nodes of the tree that predict outcomes.
• Information Gain/Gini Index: Measuring the purity (or impurity) and selecting the best attribute to split data further.
• Random Forest: A large number of decision trees that improve prediction accuracy and minimize overfitting.

Unsupervised Learning

• Clustering: Grouping similar data points into clusters based on their similarity, without predefined outputs.
• Distance/Similarity Measures: (e.g., Euclidean, Manhattan, Cosine) How to calculate similarity between objects.
• Hierarchical Clustering: Bottom up (Agglomerative, where similar items are grouped together into larger groups) or Top down (Divisive, where an entity is broken into subgroups) techniques.
• K-Means Clustering: Assigns data points to clusters based on distances to centroids.
• Association Rule Mining: Identifying co-occurrence patterns among items (e.g., customers buying bread are also likely to buy milk). It also determines the support and confidence of these associations.

Evaluation Metrics

• Accuracy: Proportion of correctly classified instances.
• Precision: Proportion of positive predictions that are actually positive.
• Recall: Proportion of actual positives that were correctly identified.
• F1-score: Harmonic mean of precision and recall.
• Mean Absolute Error (MAE): Average absolute difference between actual and predicted values.
• Mean Squared Error (MSE): Average squared difference between actual and predicted values.
• Root Mean Squared Error (RMSE): Square root of MSE.