Skip to main content

Über dieses Buch

This book covers theoretical aspects as well as recent innovative applications of Artificial Neural networks (ANNs) in natural, environmental, biological, social, industrial and automated systems.

It presents recent results of ANNs in modelling small, large and complex systems under three categories, namely, 1) Networks, Structure Optimisation, Robustness and Stochasticity 2) Advances in Modelling Biological and Environmental Systems and 3) Advances in Modelling Social and Economic Systems. The book aims at serving undergraduates, postgraduates and researchers in ANN computational modelling.



Artificial Neural Network Modelling: An Introduction

While scientists from different disciplines, such as neuroscience, medicine and high performance computing, eagerly attempt to understand how the human brain functioning happens, Knowledge Engineers in computing have been successful in making use of the brain models thus far discovered to introduce heuristics into computational algorithmic modelling. Gaining further understanding on human brain/nerve cell anatomy, structure, and how the human brain functions, is described to be significant especially, to devise treatments for presently described as incurable brain and nervous system related diseases, such as Alzheimer’s and epilepsy. Despite some major breakthroughs seen over the last few decades neuroanatomists and neurobiologists of the medical world are yet to understand how we humans think, learn and remember, and how our cognition and behaviour are linked. In this context, the chapter outlines the most recent human brain research initiatives following which early Artificial Neural Network (ANN) architectures, components, related terms and hybrids are elaborated.
Subana Shanmuganathan

Order in the Black Box: Consistency and Robustness of Hidden Neuron Activation of Feed Forward Neural Networks and Its Use in Efficient Optimization of Network Structure

Neural networks are widely used for nonlinear pattern recognition and regression. However, they are considered as black boxes due to lack of transparency of internal workings and lack of direct relevance of its structure to the problem being addressed making it difficult to gain insights. Furthermore, structure of a neural network requires optimization which is still a challenge. Many existing structure optimization approaches require either extensive multi-stage pruning or setting subjective thresholds for pruning parameters. The knowledge of any internal consistency in the behavior of neurons could help develop simpler, systematic and more efficient approaches to optimise network structure. This chapter addresses in detail the issue of internal consistency in relation to redundancy and robustness of network structure of feed forward networks (3-layer) that are widely used for nonlinear regression. It first investigates if there is a recognizable consistency in neuron activation patterns under all conditions of network operation such as noise and initial weights. If such consistency exists, it points to a recognizable optimum network structure for given data. The results show that such pattern does exist and it is most clearly evident not at the level of hidden neuron activation but hidden neuron input to the output neuron (i.e., weighted hidden neuron activation). It is shown that when a network has more than the optimum number of hidden neurons, the redundant neurons form clearly distinguishable correlated patterns of their weighted outputs. This correlation structure is exploited to extract the required number of neurons using correlation distance based self organising maps that are clustered using Ward clustering that optimally cluster correlated weighted hidden neuron activity patterns without any user defined criteria or thresholds, thus automatically optimizing network structure in one step. The number of Ward clusters on the SOM is the required optimum number of neurons. The SOM/Ward based optimum network is compared with that obtained using two documented pruning methods: optimal brain damage and variance nullity measure to show the efficacy of the correlation approach in providing equivalent results. Also, the robustness of the network with optimum structure is tested against perturbation of weights and confidence intervals for weights are illustrated. Finally, the approach is tested on two practical problems involving a breast cancer diagnostic system and river flow forecasting.
Sandhya Samarasinghe

Artificial Neural Networks as Models of Robustness in Development and Regeneration: Stability of Memory During Morphological Remodeling

Artificial neural networks are both a well-established tool in machine learning and a mathematical model of distributed information processing. Developmental and regenerative biology is in desperate need of conceptual models to explain how some species retain memories despite drastic reorganization, remodeling, or regeneration of the brain. Here, we formalize a method of artificial neural network perturbation and quantitatively analyze memory persistence during different types of topology change. We introduce this system as a computational model of the complex information processing mechanisms that allow memories to persist during significant cellular and morphological turnover in the brain. We found that perturbations in artificial neural networks have a general negative effect on the preservation of memory, but that the removal of neurons with different firing patterns can effectively minimize this memory loss. The training algorithms employed and the difficulty of the pattern recognition problem tested are key factors determining the impact of perturbations. The results show that certain perturbations, such as neuron splitting and scaling, can achieve memory persistence by functional recovery of lost patterning information. The study of models integrating both growth and reduction, combined with distributed information processing is an essential first step for a computational theory of pattern formation, plasticity, and robustness.
Jennifer Hammelman, Daniel Lobo, Michael Levin

A Structure Optimization Algorithm of Neural Networks for Pattern Learning from Educational Data

Digital technology integration is recognized as an important component in education reformation. Learning patterns of educators’ and students’ perceptions of, beliefs about and experiences in using digital technologies through self-reported questionnaire data is straightforward but difficult, due to the huge-volume, diversified and uncertain data. This chapter demonstrates the use of fuzzy concept representation and neural network to identify unique patterns via questionnaire questions. Fuzzy concept representation is used to quantify survey response and reform response using linguistic expression; while neural network is trained to learn the complex pattern among questionnaire data. Furthermore, to improve the learning performance of the neural network, a novel structure optimization algorithm based on sparse representation is introduced. The proposed algorithm minimizes the residual output error by selecting important neuron connection (weights) from the original structure. The efficiency of the proposed work is evaluated using a state-level student survey. Experimental results show that the proposed algorithm performs favorably compared to traditional approaches.
Jie Yang, Jun Ma, Sarah K. Howard

Stochastic Neural Networks for Modelling Random Processes from Observed Data

Most Artificial Neural Networks that are widely used today focus on approximating deterministic input-output mapping of nonlinear phenomena, and therefore, they can be well trained to represent the average behaviour of a nonlinear system. However, most natural phenomena are not only nonlinear but also highly variable. Deterministic neural networks do not adequately represent the variability observed in the natural settings of a system and therefore cannot capture the complexity of the whole system behaviour that is characterised by noise. This chapter implements a class of neural networks named Stochastic Neural Networks (SNNs) to simulate internal stochastic properties of natural and biological systems. Developing a suitable mathematical model for SNNs is based on the canonical representation of stochastic processes by means of Karhunen-Loève Theorem. In the implementation of this mathematical formulation for modelling nonlinear random processes from observed data, SNN is represented as a network of embedded deterministic neural networks, each representing a significant eigenfunction characterised by data, juxtaposed with random noise represented by White noise characterised by the corresponding eigenvalues. Two successful examples, including one from biology, are presented in the chapter to confirm the validity of the proposed SNN. Furthermore, analysis of internal working of SNNs provides an in-depth view of how SNNs work giving meaningful insights.
Hong Ling, Sandhya Samarasinghe, Don Kulasiri

Curvelet Interaction with Artificial Neural Networks

Modeling helps simulate the behavior of a system for a variety of initial conditions, excitations and systems configurations, and that the quality and the degree of the approximation of the models are determined and validated against experimental measurements. Neural networks are very sophisticated techniques capable of modeling extremely complex systems employed in statistics, cognitive psychology and artificial intelligence. In particular, neural networks that emulate the central nervous system form an important part of theoretical and computational neuroscience. Further, since graphs are the abstract representation of the neural networks; graph analysis has been widely used in the study of neural networks. This approach has given rise to a new representation of neural networks, called Graph neural networks. In signal processing, wherein improving the quality of noisy signals and enhancing the performance of the captured signals are the main concerns, graph neural networks have been used quite effectively. Until recently, wavelet transform techniques had been used in signal processing problems. However, due to their limitations of orientation selectivity, wavelets fail to represent changing geometric features of the signal along edges effectively. A newly devised curvelet transform, on the contrary, exhibits good reconstruction of the edge data; it can be robustly used in signal processing involving higher dimensional signals. In this chapter, a generalized signal denoising technique is devised employing graph neural networks in combination with curvelet transform. The experimental results show that the proposed model produces better results adjudged in terms of performance indicators.
Bharat Bhosale

Hybrid Wavelet Neural Network Approach

Application of Wavelet transformation (WT) has been found effective in dealing with the issue of non-stationary data. WT is a mathematical tool that improves the performance of Artificial Neural Network (ANN) models by simultaneously considering both the spectral and the temporal information contained in the input data. WT decomposes the main time series data into its sub-components. ANN models developed using input data processed by the WT instead of using data in its raw form are known as hybrid wavelet models. The hybrid wavelet data driven models, using multi-scale input data, results in improved performance by capturing useful information concealed in the main time series data in its raw form. This chapter will cover theoretical as well as practical applications of hybrid wavelet neural network models in hydrology.
Muhammad Shoaib, Asaad Y. Shamseldin, Bruce W. Melville, Mudasser Muneer Khan

Quantification of Prediction Uncertainty in Artificial Neural Network Models

The research towards improving the prediction and forecasting of artificial neural network (ANN) based models has gained significant interest while solving various engineering problems. Consequently, different approaches for the development of ANN models have been proposed. However, the point estimation of ANN forecasts seldom explains the actual mechanism that brings the relationship among modeled variables. This raises the question on the model output while making decisions due to the inherent variability or uncertainty associated. The standard procedure though available for the quantification of uncertainty, their applications in ANN model are still limited. In this chapter, commonly employed uncertainty methods such as bootstrap and Bayesian are applied in ANN and demonstrated through a case example of flood forecasting models. It also discusses the merits and limitations of bootstrap ANN (BTANN) and Bayesian ANN (BANN) models in terms of convergence of parameter and quality of prediction interval evaluated using uncertainty indices.
K. S. Kasiviswanathan, K. P. Sudheer, Jianxun He

Classifying Calpain Inhibitors for the Treatment of Cataracts: A Self Organising Map (SOM) ANN/KM Approach in Drug Discovery

Calpain inhibitors are possible therapeutic agents in the treatment of cataracts. These covalent inhibitors contain an electrophilic anchor (“warhead”), an aldehyde that reacts with the active site cysteine. Whilst high throughput docking of such ligands into high resolution protein structures (e.g. calpain) is a standard computational approach in drug discovery, there is no docking program that consistently achieves low rates of both false positives (FPs) and negatives (FNs) for ligands that react covalently (via irreversible interactions) with the target protein. Schroedinger’s GLIDE score, widely used to screen ligand libraries, is known to give high false classification, however a two-level Self Organizing Map (SOM) artificial neural network (ANN) algorithm, with KM clustering proved that the addition of two structural components of the calpain molecule, number hydrogen bonds and warhead distance, combined with GLIDE score (or its partial energy subcomponents) provide a superior predictor set for classification of true molecular binding strength (IC50). SOM ANN/KM significantly reduced the number of FNs by 64 % and FPs by 26 %, compared to the glide score alone. FPs were shown to be mostly esters and amides plus alcohols and non-classical, and FNs mainly aldehydes and ketones, masked aldehydes and ketones and Michael.
I. L. Hudson, S. Y. Leemaqz, A. T. Neffe, A. D. Abell

Improved Ultrasound Based Computer Aided Diagnosis System for Breast Cancer Incorporating a New Feature of Mass Central Regularity Degree (CRD)

Ultrasound is one of the most frequently used methods for early detection of breast cancer. Currently, the accuracy of Computer Aided Diagnosis (CAD) systems based on ultrasound images is about 90 % and needs further enhancement in order to save lives of the undetected. A meaningful approach to do this is to explore new and meaningful features with effective discriminating ability and incorporate them into CAD systems. Some of the most powerful features used in cancer detection are based on the gross features of mass (e.g., shape and margin) that are subjectively evaluated. Recently, from an extensive investigation of ultrasound images, we extracted an easily quantifiable and easily measurable new geometric feature related to the mass shape in ultrasound images and called it Central Regularity Degree (CRD) as an effective discriminator of breast cancer. This feature takes into account a consistent pattern of regularity of the central region of the malignant mass. To demonstrate the effect of CRD on differentiating malignant from benign masses and the potential improvement to the diagnostic accuracy of breast cancer using ultrasound, this chapter evaluates the diagnostic accuracy of different classifiers when the CRD was added to five powerful mass features obtained from previous studies including one geometric feature: Depth-Width ratio (DW); two morphological features: shape and margin; blood flow and age. Feed forward Artificial Neural Networks (ANN) with structure optimized by SOM/Ward clustering of correlated weighted hidden neuron activation, K-Nearest Neighbour (KNN), Nearest Centroid and Linear Discriminant Analysis (LDA) were employed for classification and evaluation. Ninety nine breast sonograms—46 malignant and 53 benign- were evaluated. The results reveal that CRD is an effective feature discriminating between malignant and benign cases leading to improved accuracy of diagnosis of breast cancer. The best results were obtained by ANN where accuracy for training and testing using all features except CRD was 100 and 81.8 %, respectively, and 100 and 95.45 % using all features. Therefore, the overall improvement by adding CRD was about 14 %, a significant improvement.
Ali Al-Yousef, Sandhya Samarasinghe

SOM Clustering and Modelling of Australian Railway Drivers’ Sleep, Wake, Duty Profiles

Two SOM ANN approaches were used in a study of Australian railway drivers (RDs) to classify RDs’ sleep/wake states and their sleep duration time series profiles over 14 days follow-up. The first approach was a feature-based SOM approach that clustered the most frequently occurring patterns of sleep. The second created RD networks of sleep/wake/duty/break feature parameter vectors of between-states transition probabilities via a multivariate extension of the mixture transition distribution (MTD) model, accommodating covariate interactions. SOM/ANN found 4 clusters of RDs whose sleep profiles differed significantly. Generalised Additive Models for Location, Scale and Shape of the 2 sleep outcomes confirmed that break and sleep onset times, break duration and hours to next duty are significant effects which operate differentially across the groups. Generally sleep increases for next duty onset between 10 am and 4 pm, and when hours since break onset exceeds 1 day. These 2 factors were significant factors determining current sleep, which have differential impacts across the clusters. Some drivers groups catch up sleep after the night shift, while others do so before the night shift. Sleep is governed by the RD’s anticipatory behaviour of next scheduled duty onset and hours since break onset, and driver experience, age and domestic scenario. This has clear health and safety implications for the rail industry.
Irene L. Hudson, Shalem Y. Leemaqz, Susan W. Kim, David Darwent, Greg Roach, Drew Dawson

A Neural Approach to Electricity Demand Forecasting

Electricity demand forecasting is significant in supply-demand management, service provisioning, and quality. This chapter introduces a short-term load forecasting model using Fuzzy Cognitive Map, a popular neural computation technique. The historic data of intraday load levels are mapped to network nodes while a differential Hebbian technique is used to train the network’s adjacency matrix. The inferred knowledge over weekly training window is then used for demand projection with Mean Absolute Percentage Error (MAPE) of 5.87 % for 12 h lead time, and 8.32 % for 24 h lead time. A Principal Component Analysis is also discussed to extend the model for training using big data, and to facilitate long-term load forecasting.
Omid Motlagh, George Grozev, Elpiniki I. Papageorgiou

Development of Artificial Intelligence Based Regional Flood Estimation Techniques for Eastern Australia

This chapter focuses on the development of artificial intelligence based regional flood frequency analysis (RFFA) techniques for Eastern Australia. The techniques considered in this study include artificial neural network (ANN), genetic algorithm based artificial neural network (GAANN), gene-expression programing (GEP) and co-active neuro fuzzy inference system (CANFIS). This uses data from 452 small to medium sized catchments from Eastern Australia. In the development/training of the artificial intelligence based RFFA models, the selected 452 catchments are divided into two groups: (i) training data set, consisting of 362 catchments; and (ii) validation data set, consisting of 90 catchments. It has been shown that in the training of the four artificial intelligence based RFFA models, no model performs the best across all the considered six average recurrence intervals (ARIs) for all the adopted statistical criteria. Overall, the ANN based RFFA model is found to outperform the other three models in the training. Based on an independent validation, the median relative error values for the ANN based RFFA model are found to be in the range of 35–44 % for eastern Australia. The results show that ANN based RFFA model is applicable to eastern Australia.
Kashif Aziz, Ataur Rahman, Asaad Shamseldin

Artificial Neural Networks in Precipitation Nowcasting: An Australian Case Study

Accurate prediction of precipitation is beneficial to many aspects of modern society, such as emergency planning, farming, and public weather forecasting. Prediction on the scale of several kilometres over forecast horizons of 0–6 h (nowcasting) is extrapolated from current weather conditions using radar and satellite observations. However, in Australia, the use of radar for nowcasting is challenging due to sparse radar coverage, particularly in regional areas. Satellite-based methods of precipitation estimation are therefore an appealing alternative; however, the ever-increasing spatial and temporal resolution of satellite data prompts investigation into options that can meet operational performance needs while also managing the large volume of data. In this chapter, the use of Artificial Neural Networks to nowcast precipitation in Australia is explored, and the current limitations of this technique are discussed. The Artificial Neural Network in this study is found to be capable of meeting or exceeding the performance of the industry-standard Hydro-Estimator method using a variety of Machine Learning metrics for the chosen verification scene. Further research is required to determine the optimal configuration of model parameters and generalisation of the model to different times and areas. This may assist Artificial Neural Networks to better reflect seasonal and orographic influences, and to meet operational performance benchmarks.
Benjamin J. E. Schroeter

Construction of PMx Concentration Surfaces Using Neural Evolutionary Fuzzy Models of Type Semi Physical Class

Pollution by particulate matter (PMx) is the accumulation of tiny particles in the atmosphere due to natural or anthropogenic activities. Particulate matter becomes a pollutant that seriously affects the health of people. In order to reduce its concentration (PMx), understanding its behavior in space is necessary, overcoming both physical and mathematical limitations. Limitations here refer to little information that a set of monitoring stations provided with regard to air quality and with respect to the dynamics of a pollutant. Furthermore, to the effect that an emission source produces within a certain area (source apportionment). Therefore, this work proposes the development of a model for spatial analytical representation of PMx concentration over time as fuzzy information. The design of the model is inspired by the structure of a Self-Organizing Map (SOM). The model consists of an input layer (n_sources) and an output layer (m_stations) that were determined in shape and size for the study area. Connections between layers are defined by a Lagrangian backward Gaussian puff tracking model, which depend on the meteorological dynamics of the area. The model allows the estimation of emissions in n_sources, based on the measurement of (PMx) concentration in the m_stations that were considered. The connection weights are adjusted by using evolutionary algorithms. The model showed a series of analytical forecasting maps that describe the spatial temporal behavior of PMx concentration in terms of the puffs emitted by n_sources. The result is a spatial neural evolutionary fuzzy model of type semi-physical class. Its application can support the improvement of air quality in an study area.
Alejandro Peña, Jesús Antonio Hernández

Application of Artificial Neural Network in Social Media Data Analysis: A Case of Lodging Business in Philadelphia

Artificial Neural Network (ANN) is an area of extensive research. The ANN has been shown to have utility in a wide range of applications. In this chapter, we demonstrate practical applications of ANN in analyzing social media data in order to gain insight into competitive analysis in the field tourism. We have leveraged the use of an ANN architecture in creating a Self-Organizing Map (SOM) to cluster all the textual conversational topics being shared through thousands of management tweets of more than ten upper class hotels in Philadelphia. By doing so, we are able not only to picture the overall strategies being practiced by those hotels, but also to indicate the differences in approaching online media among them through very lucid and informative presentations. We also carry out predictive analysis as an effort to forecast the occupancy rate of luxury and upper upscale group of hotels in Philadelphia by implementing Neural Network based time series analysis with Twitter data and Google Trend as overlay data. As a result, hotel managers can take into account which events in the life of the city will have deepest impact. In short, with the use of ANN and other complementary tools, it becomes possible for hotel and tourism managers to monitor the real-time flow of social media data in order to conduct competitive analysis over very short timeframes.
Thai Le, Phillip Pardo, William Claster

Sentiment Analysis on Morphologically Rich Languages: An Artificial Neural Network (ANN) Approach

The extraction and analysis of human feelings, emotions and experiences contained in a text is commonly known as sentiment analysis and opinion mining. This research domain has several challenging tasks as well as commercial interest. The major tasks in the area of study are, identifying the subjectivity of the opinionated sentence or clause of the sentence and then classifying the opinionated text as positive or negative. In this chapter we present an investigation of machine learning approaches mainly the application of an artificial neural network (ANN) to classifying sentiments of reader reviews on news articles written in Sinhala, one of the morphologically rich languages in Asia. Sentiment analysis provides the polarity of a comment suggesting the reader’s view on a topic. We trained from a set of reader comments which were manually annotated as positive or negative and then evaluated the ANN architectures for their ability to classify new comments. The primary interest in this experiment was the exploration of selecting appropriate Adjectives and Adverbs for the classification of sentiment in a given language. The experiment was conducted in different weighting schemes by examining binary features to complex weightings for generating the polarity scores of adjectives and adverbs. We trained and evaluated several ANN architectures with supervised learning for sentiment classification. A number of problems had to be dealt with in this experiment and they are: the unavailability of the main part of speech, adjective and adverb and the sample size of the training set. Despite the issues, our approach achieved significant results for sentence level sentiment prediction in both positive and negative classification.
Nishantha Medagoda

Predicting Stock Price Movements with News Sentiment: An Artificial Neural Network Approach

Behavioural finance suggests that emotions, moods and sentiments in response to news play a significant role in the decision-making process of investors. In particular, research in behavioural finance apparently indicates that news sentiment is significantly related to stock price movements. Using news sentiment analytics from the unique database RavenPack Dow Jones News Analytics, this study develops an Artificial Neural Network (ANN) model to predict the stock price movements of Google Inc. (NASDAQ:GOOG) and test its potential profitability with out-of-sample prediction.
Kin-Yip Ho, Wanbin (Walter) Wang

Modelling Mode Choice of Individual in Linked Trips with Artificial Neural Networks and Fuzzy Representation

Traditional mode choice models consider travel modes of an individual in a consecutive trip to be independent. However, a persons choice of the travel mode of a trip is likely to be affected by the mode choice of the previous trips, particularly when it comes to car driving. Furthermore, traditional travel mode choice models involve discrete choice models, which are largely derived from expert knowledge, to build rules or heuristics. Their approach relies heavily on a predefined specific model structure (utility model) and constraining it to hold across an entire series of historical observations. These studies also assumed that the travel diaries of individuals in travel survey data is complete, which seldom occurs. Therefore, in this chapter, we propose a data-driven methodology with artificial neural networks (ANNs) and fuzzy sets (to better represent historical knowledge in an intuitive way) to model travel mode choices. The proposed methodology models and analyses travel mode choice of an individual trip and its influence on consecutive trips of individuals. The methodology is tested using the Household Travel Survey (HTS) data of Sydney metropolitan area and its performance is compared with the state-of-the-art approaches such as decision trees. Experimental results indicate that the proposed methodology with ANN and fuzzy sets can effectively improve the accuracy of travel mode choice prediction.
Nagesh Shukla, Jun Ma, Rohan Wickramasuriya, Nam Huynh, Pascal Perez

Artificial Neural Network (ANN) Pricing Model for Natural Rubber Products Based on Climate Dependencies

International Rubber Study Group report in [1] points out that the world natural rubber consumption continues to increase at an average of 9 per cent per year. Especially, the demands of natural rubber tire industry in developed countries such as the USA, Germany, China and Japan have increased steadily. Tropical countries, such as Indonesia, Malaysia, Thailand and Vietnam, members of the Association of Natural Rubber Producing Countries (ANRPC) accounted for about 92 per cent of the global production of natural rubber in 2010. The market price of natural rubber fluctuates reflecting the variations in supply capacity of these production countries. Therefore, knowledge on the natural rubber supply from these countries is significant in order to have an accurate pricing model of natural rubbers. Moreover, the supply of natural rubber is determined by the climatic conditions in these countries. Rubber trees grow and produce best in warm with an ideal temperature between 21–35 oC, an annual rainfall of 200-300 cm and moistly conditions. In this context, the chapter looks at the dependencies of natural rubber market price especially, the climatic conditions in the production countries, and derives at a natural rubber pricing model to provide farmer information regarding the prediction of market price using an artificial neural network (ANN) based prediction approach.
Reza Septiawan, Arief Rufiyanto, Sardjono Trihatmo, Budi Sulistya, Erik Madyo Putro, Subana Shanmuganathan

A Hybrid Artificial Neural Network (ANN) Approach to Spatial and Non-spatial Attribute Data Mining: A Case Study Experience

A hybrid artificial neural network (ANN) approach consisting of self-organising map (SOM) and machine learning techniques (top-down induction decision tree/TDIDT) to characterising land areas of interest is investigated using New Zealand’s grape wine regions as a case study. The SOM technique is used for clustering map image pixels meanwhile, the TDIDT is used for extracting knowledge from SOM cluster membership. The contemporary methods used for such integrated analysis of both spatial and non-spatial data incorporated into a geographical information system (GIS), are summarised. Recent approaches to characterise wine regions (viticulture zoning) are based on either a single or composite (multi-attribute) index, formulated generally using digital data (vector and raster) representing the variability in environmental and viticulture related factors(wine label ratings and price range) over different spatial and temporal scales. Meanwhile, the world’s current wine regions, already well-developed, were initially articulated based on either grapevine growth phenology (growing degree days/GDD, frost days, average/minimum temperature, berry ripening temperature range) or wine style/rating/taste attributes. For both approaches, comprehensive knowledge on local viticulture, land area, wine quality and taste attributes is a sine qua non. It makes the characterisation of newworld vineyards or new sites (potential vineyards), with insufficient knowledge on local viticulture/environment an impossible task. For such instances and in other similar not so well-known domains, the SOM-TDIDT approach provides a means to select ideal features (discerning attributes) for characterising, in this case, within New Zealand’s wine regions or even within vineyards also scientifically validating the currently used factors regardless of present day scale and resolution related issues.
Subana Shanmuganathan
Weitere Informationen

Premium Partner