Syllabus Point
- Investigate common applications of key ML algorithms
Including:
- data analysis and forecasting
- virtual personal assistants
- image recognition
Investigate common applications of key ML algorithms
Variables in Machine Learning
In the context of machine learning, a variable is a characteristic, property or feature that can change and is used to represent data. They provide the information that the models use to learn and make predictions.
Types of Variables
- Features: independent input variables used by the ML model to make predictions
- Targets: dependent output variables that the model is trying to predict - what you want to learn about based on the features
- Labels: previously known as true values of a target variable; used during training and testing of supervised ML systems
Example: Student Test Performance Prediction
Imagine you are building a machine learning model to predict whether a student will pass or fail a test based on their study habits.
- Features: study hours per week, attendance rate, previous test scores, participation
- Target: Pass or Fail
Data Analysis and Forecasting
Data analysis involves examining large datasets to identify patterns, trends and insights. Forecasting uses these insights to predict future outcomes, by applying regression and time-series analysis algorithms.
ML Algorithms Commonly Used
- Regression algorithms: Used to forecast numeric outcomes (continuous value) such as sales, stock prices or climate patterns; finds a straight line relationship between input features and output values
- Time series forecasting: Regression model often used in sequence data; forecasts future values based on patterns in past data over time; ARIMA models, LSTM neural networks are employed for analysing trends over time
- Clustering algorithms: Group data points with similar characteristics → useful in market segmentation or anomaly detection
- Decision trees: Classify or predict outcomes based on input features
Applications
- Retail: forecasting inventory demand
- Finance: predicting stock prices, credit scoring, fraud detection
- Health: predicting disease outbreaks or patient readmission rates
Statistical Analysis
- Moving averages: Smooth out short-term fluctuations to reveal a long term trend
- Regression lines (line of best fit): Can see if there is a relationship, and whether the trend is increasing, decreasing, etc
- Seasonal decomposition: Breaks down a time series into its trend, seasonal and residual components; makes it easier to analyse seasonal variations separate from underlying trend
Types of Trends
- Linear Trends: Straight line data; consistent rate of change - easy to model and predict
- Exponential Trends: Rate of change that increases over time (e.g. population growth, technology adoption, virality); also exponential decay
- Cyclical Trends: Patterns that repeat over long periods of time (e.g. economic cycles, stock market fluctuations, real estate market); harder to predict - depend on multiple external factors
- Seasonal Trend: Repetitive pattern that occurs within a fixed period; predictable and linked to environmental, cultural or economic factors
Predictive Analysis
Predictive analytics leverages patterns in past data to make informed predictions - using data mining, statistical modelling, ML and big data analytics.
Key Techniques
- Regression analysis: Statistical method - predict a numerical outcome based on past data
- Time series forecasting: Predict future values based on sequential historical data points
- Classification models: Categorise data into predefined groups based on patterns and characteristics
- Clustering algorithms: Unsupervised learning technique (like classification, but does not require predefined labels); finds natural groupings within data
Virtual Personal Assistants
These are software agents that use natural language processing and ML to interact with users and perform tasks (e.g. Siri, Google Assistant).
Key Features
- They learn user preferences and improve interactions over time
ML Techniques Involved
- Natural language processing (NLP) for interpreting user speech or text
- Classification algorithms to categorise user intents
- Reinforcement learning to personalise responses based on user feedback and behaviour
Key Algorithms
- Neural networks: for language processing, predictive learning
- Intent recognition models: classify user intent based on spoken or typed queries
Image Recognition
Ability of software to identify objects, places, people or actions in images. Used in biometrics, security and health.
How It Works
Interpret visual data, using:
- Convolutional Neural Networks (CNNs): Effective for tasks like object detection, face recognition and medical imaging analysis
- Transfer learning: Pre-trained models are adapted for specific tasks with limited data
Applications
- Security (facial recognition)
- Health (analysing medical images like MRIs)
- Retail (image based search and product tagging)
Algorithms Used in Image Recognition
- Convolutional Neural Networks (CNNs): Automatically learn spatial features from image pixels. Most commonly used.
- K-Nearest Neighbour (KNN): Used in simpler systems to compare image feature similarity.
- Decision Trees: Can be used with pre-extracted features for basic classification tasks.
- Logistic Regression: Used for binary image classification (e.g. face or no face).
- Neural Networks: General class used in object detection, facial recognition, handwriting.
Related Resources
Keep Progressing
Use the lesson navigation below to move through the module sequence.