Business Intelligence

Post 1:

What is the relationship between Naïve Bayes and Bayesian networks? What is the process of developing a Bayesian networks model?

Naïve Bayes is a simple probability-based classification method (a machine learning technique applied to classification type prediction problems), which is derived from the famous Bayes theorem. The Bayesian network (BN) supports the self-activation and multi-directional propagation of evidence. These evidences quickly converge to a globally consistent balance. BN is a powerful tool for expressing dependency structures in a graphical, clear and intuitive way. It reflects the various states of the multivariate model and their probability relationships. Bayesian Networks can be created automatically (learnt) by using statistical data examples (Zdravko & Daniel, 2017).

Naïve Bayes advantages include the ability to develop efficiently in a machine learning environment. Bayesian networks advantage is its adaptability, which can start to build a network with limited knowledge of the model and expand as new information is obtained. In addition, the method has good applicability because the complete BN provides a holistic view of all relationships. The relationship between Naïve Bayes and Bayesian networks are a Bayesian Networks does not assume independence among the input variables (Sharda, R., Delen, D., Turban, E, 2020).

Manual Construction (Directed acyclic path and Conditional probability distribution) and Automatic learning are the methods used to develop a Bayesian networks model. For Manual Construction all the conditional probability distributions are assumed to be prior known. Bayesian network can automatically learn directly from the database using experience-based algorithms that are usually built into the appropriate software (Michal Horny, 2014).

The graph model uses a conditional probability distribution on each node of the graph. If the conditional probability distribution is unknown, you can obtain it from the data by estimating the empirical conditional probability distribution (conditional frequency). In the case of automatic learning, all relevant variables must be organized in a database Structure.

References:

Zdravko, M. and Daniel T. (2017). Data Mining the Web: Uncovering Patterns in Web Content, Structure, and Usage, Wiley.

Sharda, R., Delen, D., Turban, E. (2020). Analytics, Data Science, & Artificial Intelligence: Systems for Decision Support 11E.

Michal, H. (April 18, 2014). https://www.bu.edu/sph/files/2014/05/bayesian-networks-final.pdf

Post 2

The strong relationship between Bayesian networks and Naive Bayes is explained by the fact that in Naive Bayes, the positive predictive value of each observation, and the probability of seeing that observation again in the next training dataset, is one nothing can give new predictions. However, Naive Bayes is a stochastic process, and so can have Gaussian distributions and error functions. The independence of the outputs of the Gaussian process with the outputs of the stochastic process is equal, but Naive Bayes has a strict independence requirement. can use the independence of the output of a stochastic process, namely a Gaussian distribution as a conditional predictor (Arnott, Lizama, & Song, 2017)

Bayesian networks are models of data distribution. In all cases, their distribution can be written in terms of stochastic variables that can be drawn from a uniform distribution. One can also argue that sequential, random variates, collected from the samples concerning the template, would constitute a continuous distribution. The process for developing Bayesian network models follows a well-defined procedure. Another way of designing a Bayesian network model is to consider using an objective function that says that the data was generated under a given parameter (Arnott, Lizama, & Song, 2017).

To create a Bayesian network model. For example, if have 400 features, it will have several parameters for which have to specify the probability of the given feature being true. Building a Bayesian network model can take at least a few hours. As the course progresses, you'll learn about a number of advanced statistical techniques that can improve your model evaluation and performance There are various methods for modelling these models, but the approach described in this article is the one that I personally find best, to satisfy the objectives (Yeoh, & Popovic, 2016).

 References

Arnott, D., Lizama, F., & Song, Y. (2017). Patterns of business intelligence systems use in organizations. Decision Support Systems, 97, 58-68.

Yeoh, W., & Popovič, A. (2016). Extending the understanding of critical success factors for implementing business intelligence systems. Journal of the Association for Information Science and Technology, 67(1), 134-147.