Artificial Intelligence

Artificial Intelligence is an important scientific area which is expanding in various applications and is experiencing significant renaissance of optimism. In MatDeck, several artificial intelligence algorithms have been implemented in two broad areas, at this moment in time. The first area is time series processing with a heavy focus on forecasting future values of economic time series, and the second area is based on classification and machine learning.

The traditional approaches to time series prediction, especially for economic series, are Box-Jenkins or ARIMA methods. The original ARIMA model uses an iterative three-stage modeling approach which consists of: model identification and selection, parameter estimation, and model realization. In MatDeck, autocorrelation and partial autocorrelation function – pacf() are used for model identification. For parameter estimation, maximum likelihood estimation computation algorithms are used to arrive at coefficients, for example Yule-Walker estimation – yulewalker() is used to estimates AR parameters. Forecasting is implemented using appropriate models with estimated parameters – arforecasting(). The quality of predictive capabilities in MatDeck can be tested and evaluated using various statistical error functions such as mean square error – mse(), mean percentage error – mpe() etc.

The Naive Bayes algorithm is one of the very first and effective methods used for classification and machine learning which is implemented in MatDeck. The Naive Bayes algorithm is a clever method that makes an informed decision regarding the classification of something by using the probabilities of all the attributes (features) belonging to each class. It greatly simplifies the process of calculating the probabilities of each attribute by assuming that the probability of each attribute in a given class is independent of all other attributes. In MatDeck, the implementation of Naive Bayes consists of two functions. The first function, naivebayest(), is used for training and  prepares the data in a format to be used in the Naive Bayes classification function. The second function, naivebayesc(), is used to calculate the posterior probabilities based on the training data in order to make an informed decision and then classify a record defined by the attributes.

There are two examples which illustrate time-series forecasting in MatDeck. For illustrating, stock price values are predicted based on past values.

There are two key examples that illustrate how the naive Bayes classifier can be utilized in MatDeck.