Abstract
Neural Networks (NN) are one of the most used machine learning techniques in different areas of knowledge. This has led to the emergence of a large number of courses of Neural Networks around the world and in areas where the users of this technique do not have a lot of programming skills. Current software that implements these elements, such as Matlab®, has a number of important limitations in teaching field. In some cases, the implementation of a MLP requires a thorough knowledge of the software and of the instructions that train and validate these systems. In other cases, the architecture of the model is fixed and they do not allow an automatic sweep of the parameters that determine the architecture of the network. This chapter presents a teaching tool for the its use in courses about neural models that solves some of the above-mentioned limitations. This tool is based on Matlab® software.
TopBackground
In the literature it is possible to find all kinds of educational software developed for different engineering courses related to data analysis and machine learning. In (Carrasco Fernández, 2012), the authors present an educational software developed using the MATLAB GUIDE tool. This software allows engineering students gain knowledge about data sets via the exploratory data analysis (EDA). This application includes models like clustering algorithms and self-organizing maps. However, does not include multilayer perceptrons. In (Deperlioglu, 2011) an educational tool to work with different kinds of neural network models is presented. The developed tool includes MLP, LVQ and SOM models. The design of the models was done visually and interactively but, in contrast to the proposed application in this chapter, confusion matrix or sensitivity-specificity histograms are not shown in the results. Moreover, in (Marković, 2014) a software system developed to support the teaching of Intelligent Systems is presented. The tool includes decision trees (ID3), clustering (kmeans), Naive Bayes, and perceptron models. In works (Ugur, 2010; Hwang, 2003; García Roselló, 2003; Zatarain, 2011) similar applications are presented.
Key Terms in this Chapter
Sensitivity/Specificity: Sensitivity and specificity are statistical measures of the performance of a classification problem. Sensitivity measures the success in the positive class and specificity measures the success in the negative class.
Local Minimum: Is the lowest value on the nearest points in a function, but not the smallest absolute value. It is important to avoid these local minima to obtain a more accurate model.
Confusion Matrix: A table that allows the visualization of a classification algorithm performance. It is formed by two rows and two columns that reports the number of successes/failures in the positive/negative class.
Classification Problem: The objective in these problems is assigning values to one of several predefined categories.
Machine Learning: A subfield of computer science and artificial intelligence which aims to develop techniques that allow machines to solve problems automatically. This learning is done by generalizing problems based on training data.
Model Overfitting: Occurs when a model describes random error or noise instead of the underlying data relationship because the model is excessively complex, such as having too many parameters relative to the number of observations.
Training/Validation Set: The data set is divided into two subsets. The training subset is used by the algorithm to learn the problem and validation subset is used to verify that the model can solve the problem with unknown new data.
Regression Problem: The objective in these problems is to obtain predictions about future values?? obtained from current and past values.