LEARNING PATH: MATLAB: Powerful Machine Learning with MATLAB

We will explore decision trees methods. Then, we will learn the concepts like nodes, branches, and leaf nodes. We will see how to classify objects into a finite number of classes by repeatedly dividing the records into homogeneous subsets with respect to the target attribute.

We will learn the basic concepts of probability theory: classical probability definition, dependent and independent events, joint probability and conditional probability, which is the basis of these methods.

In this video we will explore discriminant analysis methodologies; several examples will be analyzed to compare the relative results. We will also learn how to create models to minimize the expected classification cost.

With nearest neighbor classifiers, we will learn how to identify the class of a sample based on the distance of it from other classified objects. Discover how to fix the distance metric and how to choose the optimal value for K. So, as to understand how to improve model performance through cross-validation.

In this video, we will analyze the Classification Learner app, and how it leads us into step-by-step classification analysis. With the help of this app, to import and explore data, select features, specify validation schemes, train models, and evaluate results, will be extremely simple and fast.

We will look at a couple of methods for grouping objects: hierarchical clustering and partitioning clustering. In the first method, clusters are constructed by recursively partitioning the instances in either a top-down or bottom-up fashion. The second one decomposes a dataset into a set of disjoint clusters.

To determine the proximity of objects to each other, we will use the linkage function. With the cluster function, we will cut the ramifications from the bottom of the hierarchy tree and assign all the objects below each cut to a single cluster.

In this video, we will explore partitioning clustering through the k-means method. Then we will learn how to locate K - centroids, one for each cluster, by an iterative procedure. We will discover how well these clusters are separated and how to make a silhouette plot using cluster indices issued by K-means.

K-medoids is more robust to noise and outliers than K-means, because a mean is easily influenced by extreme values. We will learn to use the K-medoids function; it partitions the observations of a matrix into k clusters and returns a vector containing the cluster indices of each observation.

The models are composed of k (positive integer) multivariate normal density components. Each component has an n-dimensional mean, n-by-n covariance matrix, and mixing proportion. We will use the fitgmdist function to return a Gaussian mixture distribution model with k components (fixed by the user) fitted to the input dataset.

Neural networks work in parallel and are therefore able to deal with lots of data at the same time, as opposed to serial computers, in which each one is processed individually and in succession.

Neuron input is a set of fibers called dendrites; they are in contact with the axonsof other neurons from which they receive electrical potentials. The connection point between an axon of a neuron and the dendrite of another neuron is called synapse.

The Neural Network Toolbox provides algorithms, pre-trained models, and apps to create, train, visualize, and simulate neural networks with one hidden layer and neural networks with several hidden layers. Let’s see how to do this?

To make the application of neural networks as simple as possible, the toolbox gives us a series of GUIs. In this video, We will check out the neural network getting started GUI, the starting point for our neural network fitting, pattern recognition, clustering, and time series analysis.

In this video, we will focus on fitting data with a neural network. We will see how to use the Neural Fitting app (nftool). Then, we will run a script analysis to learn how to use neural network functions from the command line.

Selection of features is necessary to create a functional model so as to achieve a reduction in cardinality, imposing a limit greater than the number of features that must be considered during its creation. Feature selection is based on finding a subset of the original variables, usually iteratively, thus detecting new combinations of variables and comparing prediction errors.

The process of transforming the input data into a set of features is called feature extraction. If the features extracted are carefully chosen, it is expected that the features set will perform the desired task using the reduced representation instead of the full size input.