⛱️Cognitive Computing in Business Unit 3 – Machine Learning Essentials

What's Machine Learning Anyway?

• Machine learning (ML) is a subset of artificial intelligence that focuses on building systems that can learn and improve from experience without being explicitly programmed
• ML algorithms build mathematical models based on sample data, known as training data, to make predictions or decisions without being explicitly programmed to do so
• As the models are exposed to new data, they adapt and learn from previous computations to produce reliable, repeatable decisions and results
• ML is closely related to computational statistics, which focuses on making predictions using computers, but not all ML is statistical learning
• The study of mathematical optimization delivers methods, theory and application domains to the field of machine learning
• ML algorithms are used in a wide variety of applications (email filtering, computer vision, recommendation engines) where it is difficult or unfeasible to develop conventional algorithms to perform the needed tasks
  • Spam filtering is a common example, where ML algorithms learn to flag spam based on words or phrases found in the email text
• ML enables analysis of massive quantities of data, delivering faster, more accurate results to identify profitable opportunities or dangerous risks

Key Machine Learning Concepts

• Dataset is a collection of data used to train and test ML models, typically split into training, validation, and test sets
• Features are the input variables used in making predictions, represented as columns in tabular datasets or certain abstract qualities (texture, color, etc.) in unstructured data like images
• Labels are the output variable or target that the model is trying to predict, based on the features
• Model is the mathematical representation of a real-world process, trained on historical data to make predictions on new data by capturing the relationship between features and label
• Training is the process of feeding data to the model so it can learn the relationships between features and label
  • The goal is to minimize the difference between the model's predictions and the actual labels, known as the loss or cost function
• Inference is when the trained model is used to make predictions on new, unseen data
• Hyperparameters are the settings used to control the model training process (learning rate, number of hidden layers in a neural network, etc.)
  • These are set before training and are not learned from data like model parameters

Types of Machine Learning

• Supervised learning uses labeled datasets to train algorithms that to classify data or predict outcomes accurately
  • Input data is called training data and has a known label or result (historical stock prices, images of dogs labeled "dog", etc.)
  • Model is trained until it can detect the underlying patterns and relationships, enabling it to yield accurate labeling results when presented with never-before-seen data
• Unsupervised learning is used on data with no historical labels, allowing the algorithm to act on that data without guidance
  • Unsupervised learning can discover hidden patterns or data groupings (customer segments) without the need for human intervention
  • Commonly used for transactional data (identifying segments of customers with similar attributes who can then be treated similarly in marketing campaigns)
• Semi-supervised learning uses a mix of labeled and unlabeled data, usually a small amount of labeled data with a large amount of unlabeled data
  • Can address real-world problems where labeled data is scarce or expensive, but unlabeled data is abundant
• Reinforcement learning trains models to make a sequence of decisions by exposing the model to the environment where it trains itself based on feedback
  • Learns by trial and error to capture the best possible knowledge to make accurate decisions
  • Often used in gaming, robotics, and navigation

The Machine Learning Process

• Define problem and gather data from various sources (databases, sensors, APIs) and different formats (tables, images, text)
• Prepare data by cleaning (handling missing values, removing duplicates), transforming (scaling, encoding categorical variables), and splitting into train, validation, and test sets
  • Exploratory data analysis examines data to understand its main characteristics (mean, standard deviation, correlation, etc.)
  • Feature engineering creates new input features from existing ones to improve model performance
• Choose a model based on the problem type (classification, regression, clustering), data size and type, and resource constraints
• Train the model on the training data, tuning hyperparameters to improve performance
  • Validation data provides an unbiased evaluation of model performance while tuning hyperparameters
• Evaluate model performance on the test set using appropriate metrics (accuracy, precision, recall, F1-score)
• Deploy model into production environment to start generating predictions on real-world data
• Monitor and maintain model performance over time, retraining or updating as needed based on new data or changing environment

Popular ML Algorithms

• Linear Regression predicts continuous values (house prices) by fitting a linear equation to observed data
• Logistic Regression predicts binary outcomes (spam or not spam) by fitting a logistic function to observed data
• Decision Trees predict outcomes by learning simple decision rules inferred from the data features
  • Random Forest is an ensemble of decision trees, making predictions by aggregating the predictions of multiple trees
• Support Vector Machines find a hyperplane in N-dimensional space (N = number of features) that distinctly classifies the data points
• Naive Bayes classifiers are a family of probabilistic algorithms based on applying Bayes' theorem with strong independence assumptions between the features
• K-Means Clustering groups unlabeled data into K clusters based on feature similarity
• Neural Networks are inspired by biological neural networks, consisting of input, hidden, and output layers of interconnected nodes
  • Deep Learning uses neural networks with many hidden layers to learn hierarchical representations of data

Tools and Frameworks

• Python is the most popular programming language for ML due to its extensive libraries (NumPy for numerical computing, Pandas for data manipulation, Matplotlib for visualization)
• Scikit-learn is a Python library that provides a wide range of supervised and unsupervised learning algorithms
• TensorFlow is an open-source library for dataflow and differentiable programming, used for ML applications such as neural networks
• Keras is a high-level neural networks API written in Python, capable of running on top of TensorFlow, Microsoft Cognitive Toolkit, R, Theano, or PlaidML
• PyTorch is an open source ML library based on Torch, used for applications such as computer vision and natural language processing, primarily developed by Facebook's AI Research lab
• Apache Spark is a fast and general-purpose cluster computing system that provides APIs to work with large datasets, including Spark MLlib for ML
• Cloud platforms like Amazon Web Services, Google Cloud, and Microsoft Azure offer managed ML services for building, training, and deploying models at scale

Real-World Business Applications

• Fraud Detection uses ML to identify suspicious patterns and prevent fraudulent transactions in industries like banking and insurance
  • JPMorgan Chase uses ML to detect fraud and money laundering, saving $150 million annually
• Recommendation Systems suggest relevant products or content to users based on past behavior and similar users' preferences
  • Netflix uses ML to personalize movie and TV recommendations, saving $1 billion annually in customer retention
• Customer Churn Prediction helps businesses identify customers at high risk of leaving, allowing proactive retention efforts
  • Verizon uses ML to predict customer churn and improve retention by 1-5%
• Dynamic Pricing optimizes product prices in real-time based on demand, competitor prices, and other market factors
  • Uber uses ML for surge pricing to balance supply and demand by raising prices when demand outstrips supply
• Predictive Maintenance anticipates equipment failures to allow maintenance to be scheduled before the failure occurs, preventing unexpected downtime
  • General Electric uses ML to predict jet engine failures, reducing flight delays and cancellations
• Image and Video Analysis extracts insights from visual data for applications like medical diagnosis, defect detection in manufacturing, and autonomous vehicles
  • IBM Watson uses ML to assist doctors in diagnosing diseases like cancer from medical images
• Natural Language Processing enables computers to understand, interpret, and generate human language for applications like sentiment analysis, chatbots, and machine translation
  • Google Translate uses ML to provide real-time language translation for over 100 languages

Challenges and Limitations

• Data quality issues like missing values, outliers, and noise can significantly impact model performance
  • Garbage in, garbage out - ML models are only as good as the data they are trained on
• Model interpretability is a challenge, particularly for complex models like deep neural networks
  • Black box models make it difficult to understand how the model arrives at its predictions, which can be problematic in regulated industries
• Bias in training data can lead to biased predictions, perpetuating or even amplifying societal biases
  • Amazon had to scrap an ML recruiting tool that showed bias against women due to historical hiring data
• Overfitting occurs when a model learns the noise in the training data to the extent that it negatively impacts the performance on new data
  • Regularization techniques (L1/L2 regularization, dropout) can help mitigate overfitting
• Deployment and scaling ML models in production environments can be challenging due to computational requirements and the need to retrain models as new data becomes available
  • Techniques like model compression and quantization can help reduce model size and inference latency
• Adversarial attacks involve malicious actors manipulating input data to fool ML models and cause misclassifications
  • Adversarial training incorporates adversarial examples in the training data to improve model robustness
• Concept drift occurs when the statistical properties of the target variable change over time, leading to a degradation in model performance
  • Continuous monitoring and retraining of models is necessary to adapt to changing environments