Predictive Modelling in Business Environment, Artificial Neural Networks, and K-Nearest Neighbor Algorithms - Essay Example

PREDICTIVE MODELING Name Institution Predictive Modelling In Business Environment With increasing levels of competition, businesses have been focusing on the use of various scientific models to improve the quality and efficiency of the decision making process. Most businesses have been trying to improve their intelligence using a set of mathematical models and methodologies in exploiting the available data and generating information that is considered crucial in decision-making process. Some of the predictive models used include the artificial neural network (ANN), support vector machines (SVM), and k-nearest neighbor (kNN) (Yang, 2010). Biologically, neural networks have been used to refer a network of biological neural; however, from the business point, the term refers to artificial neurons. The biological neural network encompasses real biological networks, which are connected in the peripheral nervous systems. Within neuroscience profession, these neutron networks perform specific physiological functions. On the other hand, the ANN composed of interconnecting artificial neurons aimed at increasing the level of understanding of the functionality of biological neural networks or solving business problems that require intelligence. Functionality of ANN ANN model has a mapping capability. As a result, they can map input patterns to the associated output patterns. Moreover, the models application learns from examples, that is, it could be trained using known examples of prevailing problems affecting business performance before their testing for inference capacity on the already known instances of the problem. Besides, the models have the ability of identifying new objects that were previously untrained since they possess the ability to generalize the inputs. Therefore, the models have the ability of predicting the new outcomes from the past encounters. The model is a robust system and highly tolerant to faults. It has the ability of recognizing incomplete patters from full patters, partial and noisy ones (Funk, 2009). More importantly, the model can process information in parallel, at a faster speed, and distributive manner. ANN borrows the functional concept of the biological neuron systems hence mimicry of the biological process. ANN receives analogous inputs to the electrochemical impulses. In biological application, the neurons function properly by ensuring that its systems are adjusted to the synaptic connection that exists between the neurons (Metz, 2015). As a result, ANN constitutes a flexible non-linear model that imitates the functions of the biological neural system. The application of the model did not receive much attention initially due to its limited functionality in computing data per unit time. Moreover, it functions were limited in terms of applications. Through the years, there have been modifications on the model including multi-layered ANN that has the ability to approximate large class of functions to the extent of any desired accuracy. The use of the model tends to differ from one organization to another including pattern recognition, language learning, and signal processing. Any ANN processing elements primarily conducts two functions: computing the output, which is sent to other elements outside the network where every neutron determines the value of its output through application of the transfer functions, and updating the local memory. The data involved in the memory include weights and variable data. These neurons are organized into layers with the first layer known as the input layer while the last is the output layer. It is important to note that there are inner layers ranging from one to many identified as the inner layer. The input neurons receive input signals from external source while the outputs send their values to the ANN. Both hidden and output neurons receive input signals from the incoming connections and signals from its local memory. The ANN functions as information-processing systems and serving as general software tool across weighted connections. The model is inspired by the architect of the human brain exhibiting features such as the ability of learning complex patterns of information. This information could be used to analyze data, prediction, classification, and clustering. Structurally, the model has several processing nodes that generally accept values from other neurons by the input arcs. Depending on the method of connection among the neurons, ANN model could be divided into two: feed forward neural networks, which only have forward information transfer but has no feedback information and feedback neural network that has both the ability of forwarding transfer of information and reversing transfer information (Kwon, 2011). Generally, feed forward neural networks are made up of an input layer, hidden layer, and an output layer. The neurons in every layer accept output information from the neurons of the forward layer while in the feed forward network; the information always moves from a single direction and never goes backwards. Figure 1: Interaction of the components of artificial neural network Figure 2: Functionality of the artificial neural network Comparison and contrasting of the artificial neural networks, support vector machines, and k-nearest neighbor algorithms All the three algorithms could be used in clarifying and predicting the expected information. All the models indicate efficacy and accuracy depending on the type of dataset used and the type of the prevailing problem. Furthermore, all the models functionalities depend on the learning process by applying different datasets. Considering SVM, it has higher capacity of generalizing datasets with smaller margins compared to the other models. Moreover, the model has the ability of using the data from the remote sensing and classifying the information in a more effective and efficient manner. Besides, SVM plays an integral part in classifying binary data (Prakash, 2012). SVM is mainly used when reducing the structural risks associated with the probability functions while both kNN and ANN focus on risk reduction in the empirical data. All the models have the ability of recognizing different patterns fed into the system using decision function. Several factors influence the performance of kNN. Since the functionality and accuracy of the model depends on the distance factor, it is important to consider the distance, the decision rules while deriving the function and the number of neighbors used in classifying new examples. Even though all the models require expertise to use, the kNN is quite simple and more often uses generalized data based on the logic reasoning. The model also requires little experience. Focusing on the level of accuracy, SVM offers data that are more accurate compared to ANN and kNN. However, it is important to note that this accuracy mainly relates to information based on maximum probability algorithm (Abd, 2012). SVM in most cases is considered superior to ANNs due to weaknesses associated with it. The convergence of ANN is local minima and not global making it lack essential data for managing larger organizations. In some cases, the ANN model often over fit especially when the training goes long; hence begin to consider noises as part of the patter. Strengths and Weaknesses of the Models Although all the models play crucial in solving most organizational problems, they experience several challenges. For example, the result from SVM analysis is purely dichotomous hence offers no probability to the data in question. The complex nature of the models in most cases becomes a barrier to many organizations planning to incorporate them into the management system since they require expertise. SVM is more complex than complex and require greater memory. As a result, its testing speed is quite slow as compared to the other models. When used as a classifier, kNN require a wider storage capacity, which at times makes it more expensive for many organizations to afford. Most organization often prefers utilizing ANN due to its strengths. These strengths include its ability of accommodating numerous data including those considered noisy. Besides, it is the model with the highest algorithms. Even with these strengths, the applications of the model involve a greater burden relating to computation. Considering its level of susceptibility to numerical errors, it requires the longest duration of training. While using kNN, to define accurately, the distance function might be a greater problem. To some extent, the neighbor might also provide information for the neighboring data during classification process, thus leading to inaccuracy of the output (Campbell & Ying, 2011). In areas where the black box models are inadequate, this weakness might be considered as the strength. Measuring the metrics measure distances existing between the data is another challenge associated with kNN. As a result, it is associated with trials and errors especially while defining the metrics of a given type of data. With such approximations, the accuracy of the output is questionable. In most cases, there are cases of manipulation of the data in this model to acquire the desired output, which contribute to adaptive behaviors. Application of the ANN Model in Business Management With the business enterprises experiencing high level of competitions, change in the market trends, and increased managerial duties, there is need to employees the use of models to help solve these problems. ANN has several applications in the world of business in solving problems. Since the model plays an important role in identifying the patterns and data trend, making it suitable for prediction or forecasting on desired outcome. Some of the areas of use include sales forecasting, customer research, risk management, data validation, target marketing, and industrial processes. Google Company has been on the lead with its software development especially the one inspired by the human brain (Simonite, 2012). This software, neural networking, employs the technique of collecting data and uses it for other processes like the neurons found in the human brain while learning something. Moreover, the company intends to avail in across for commercial purposes. Fortunately, the company has used the neural networking data in computers in recognizing cats in videos from YouTube. The computer had the ability of detecting whether the features are video based on the patterns and colors. Furthermore, the company intends to revolutionize the model to advanced speech recognition technology, especially in the Android devices. The company needs to make the model work like Apple’s Siri technology, which has the ability of recognizing the voices of people, which is also used to control the gadget. Additionally, the model is capable of gathering data using artificial intelligence and make inferences on whatever issues people are discussing in different situations. References Abd, E. M. (2012). Unconstrained facial expression recognition in still images and video sequences using Random Forest classifiers. Montreal: McGill University Libraries. Campbell, C., & Ying, Y. (2011). Learning with support vector machines. San Rafael, Calif.: Morgan & Claypool. Funk, T. (2009). Web 2.0 and beyond: Understanding the new online business models, trends, and technologies. Westport, CT: Praeger. Kwon, S. J. (2011). Artificial neural networks. New York: Nova Science Publishers. Metz, C. (2015, April 21). Finally, Neural Networks That Actually Work | WIRED. Retrieved from http://www.wired.com/2015/04/jeff-dean/ Prakash, N. (2012, October 9). Google's Neural Networks Advance Artificial Intelligence [VIDEO]. Retrieved from http://mashable.com/2012/10/09/google-artificial-intelligence/ Simonite, T. (2012, October 5). Google Puts Its Virtual Brain Technology to Work | MIT Technology Review. Retrieved from http://www.technologyreview.com/news/429442/google-puts-its-virtual-brain- technology-to-work/ Yang, Z. R. (2010). Machine learning approaches to bioinformatics. Singapore: World Scientific. Read More