Solving Supervised Learning Problems

Supervised learning problems, where an output attribute is present in the dataset and the target of the analysis is to derive a model that describes the relationship between this output attribute and other input attributes in the dataset.

Depending on the output variable type, two different supervised learning problems can be defined:

  • classification problems aim to define which class or category input attributes in a dataset belong. The output attribute that defines the target class or category is a nominal attribute. For example, a classification problem may predict whether car sales will increase, decrease or remain stable (3 possible outcomes) over the next 12 months.

  • regression problems aim to predict a continuous quantity output for a set of input attributes in a dataset. For example a regression problem may predict how much a house will be worth in twelve months from now (there are many possible outcomes, all of which are continuous values).


Rulex's Logic Learning Machine

Rulex has its own proprietary algorithm for dealing with supervised learning problems, called the Logic Learning machine (or simply LLM).

The LLM algorithm produces intelligible rules, pretty much the same way as the human brain creates predictions, using learned logic.

For example:

IF weather = raining AND meeting before 9am = FALSE, THEN Best Method of Transport = Bus
IF weather = sunny AND meeting before 9am = FALSE, THEN Best Method of Transport = Bicycle
IF weather = raining AND meeting before 9am = TRUE, THEN Best Method of Transport = Car

These human readable and clear rules are expressed as if-then rules within Rulex, and allow users to understand and explain the reasoning behind predictions.

This makes Rulex the leader in eXplainable AI (XAI) and provides an ethical and GDPR compliant AI tool.

Rulex's Hierarchical Basket Analysis task also produces these if-then human understandable rules.

The following three LLM tasks are provided to solve classification, regression and one-class problems:

Problem

Description

Corresponding page

Classification

The aim of classification problems is to define which class or category input attributes in a dataset belong.

The output attribute that defines the target class or category is a nominal attribute.

Using LLM to Solve Classification Problems

Regression

The aim of regression problems is to predict a continuous quantity output for a set of input attributes in a dataset.

Using LLM to Solve Regression Problems

One-Class

One-class classification problems, where all the data under analysis belongs to the class, is mainly used to subsequently detect anomalies or outliers.

Using LLM to Solve One-Class Problems


Rulex Classification Tasks

Apart from its proprietary LLM Classification task, Rulex offers a full range of "traditional" machine learning tasks for solving classification problems.

Task

Description

Corresponding page

Decision Tree

Solves classification problems by building a tree structure of intelligible rules.

Using Decision Tree to Solve Classification Problems

K-Nearest Neighbor

Determines the output of a new example on the basis its nearest neighbors.

Using K-Nearest Neighbor to Solve Classification Problems

Logistic

Solves classification problems according to the logistic regression approach.

Using Logistic to Solve Classification Problems

Neural Networks

Builds a feed-forward neural network for classification.

Using Neural Networks to Solve Classification Problems

SVM

Trains a Support Vector Machine.

Using SVM to Solve Classification Problems


Rulex Regression Tasks

Along with its proprietary LLM Regression task, Rulex offers a full range of "traditional" machine learning tasks for solving regression problems.

Task

Description

Corresponding page

Auto Regressive

Predicts the future values assumed by a signal over time, principally based on its past behavior.

Using Auto Regressive to Solve Regression Problems

K-Nearest Neighbor

Determines the output of a new example on the basis its nearest neighbors.

Using K-Nearest Neighbor to Solve Regression Problems

Linear

Solves regression problems in which the output value is expected to be a linear combination of the input variables through the Ordinary Least Squares (OLS) method.

Using Linear to Solve Regression Problems

Neural Networks

Builds a feed-forward neural network for regression.

Using Neural Networks to Solve Regression Problems

Regression SVM

Trains a Support Vector Machine for regression.

Using Regression SVM to Solve Regression Problems

Regression Tree

Builds a regression model using the decision tree approach.

Using Regression Tree to Solve Regression Problems