The core idea is to minimize the difference between the predicted values and the actual values, typically using methods like Ordinary Least Squares (OLS). The 'best-fit' line is determined by finding the coefficients (m and c) that minimize the sum of squared errors. In IoT, this can be used to forecast temperature based on historical data or predict the power consumption of a device based on its operational parameters.

The sigmoid function, σ(z) = 1 / (1 + e^-z), transforms a linear combination of input features (z) into a probability. A threshold (commonly 0.5) is then applied to classify the instance into one of two classes. For IoT, it's ideal for anomaly detection (e.g., detecting a faulty sensor reading) or classifying events (e.g., identifying if a motion sensor has been triggered).

The core principle is to find the hyperplane that has the largest distance to the nearest training data points of any class. This maximum margin classifier is less sensitive to outliers and generally performs well. For complex, non-linear separation, SVMs utilize kernel tricks (like polynomial or radial basis function kernels) to map data into higher dimensions where linear separation is possible. In IoT, SVMs can be used for image classification (e.g., identifying objects captured by a camera) or for classifying different types of network traffic.

The tree is built by recursively partitioning the data based on the feature that best splits the data according to a certain criterion (e.g., Gini impurity or information gain for classification, variance reduction for regression). This process continues until a stopping criterion is met (e.g., maximum depth, minimum samples per leaf). Decision Trees are prone to overfitting, so techniques like pruning or using ensemble methods (like Random Forests or Gradient Boosting) are often employed. In IoT, they can be used for simple rule-based decision making, such as determining if a device should enter a low-power mode based on sensor inputs.