Efficiency Of Artificial Neural Networks Research Articles

Efficiency investigation of artificial neural networks in human activity recognition

Nowadays, human activity recognition (HAR) is an important component of many ambient intelligent solutions where accelerometer and gyroscope signals give the information about the physical activity of an observed person. It has gained increasing attention by the availability of commercial wearable devices such as smartphones, smartwatches, etc. Previous studies have shown that HAR can be seen as a general machine learning problem with a particular data pre-processing stage. In the last decade, several researchers measured high recognition rates on public data sets with numerous “shallow” machine learning techniques. In some cases artificial neural networks (ANNs) produced better performance than other shallow techniques while in other cases it was less effective. After the appearance of deep learning, a significant part of HAR researches turned toward more complex solutions such as convolutional neural networks (CNNs) and they claimed that CNNs can substitute the feature extraction stage in shallow techniques and can outperform them. Therefore in the current state of the art, the efficiency of ANNs against CNNs and other machine learning techniques is unclear. The aim of this study is to investigate the performance of more ANN structures with different hyper-parameters and inputs on two public databases. The result will show that the two key factors in ANN design are the data-preprocessing and the hyper-parameter setup because the accuracy difference between a well and a badly parameterized ANN is huge. A well-tuned ANN with extracted features can outperform other machine learning methods in the HAR problem including CNN classifiers.

English

In this study, the physical and chemical characteristics of Bambusa vulgaris ex J.C. Wendl. var.vulgaris (Bambusa vulgaris) aged 1, 2 and 3 years were evaluated. The objective was to train, validate and evaluate the efficiency of artificial neural networks (ANNs) as predictive tools to estimate bamboo stem energy density grown commercially in northeastern Brazil. For that, samples were collected in a commercial plantation and managed for energy production, determining the energy properties. Among all the characteristics analyzed, basic apparent density was the one with major correlation with bamboo stem energy density. This factor has a great advantage because it is easy to estimate, determined both by dry mass at 0% moisture, and at saturated mass. Also, the precision of ANNs was verified when associated with basic density, as a predictor of bamboo stem energy density, showing low standard error (Syx%, 1.52) and high coefficient of determination (R² = 0.98). ANN-estimated values had no statistical difference (tcal 0.58 ≤ ttab 2.08) with energy density estimated in the laboratory. Therefore, this tool was efficient, being recommended to predict the energetic density of the species under study, with basic density as the only predictive variable.   Key words: Bamboo, biomass, energy, artificial intelligence.

Eficiência da estimação da área foliar de couve por meio de redes neurais artificiais

RESUMO A estimativa da área foliar na couve é importante, pois medidas diretas são difíceis e imprecisas, devido ao tamanho da folha, a irregularidade da superfície foliar de alguns genótipos, a necessidade de equipamentos caros e de muita mão-de-obra. Objetivou-se verificar a eficiência da estimação da área foliar de couve por meio de RNAs e constatar a eficiência desta estratégia em comparação com o uso da área foliar observada. O experimento foi conduzido em delineamento de blocos casualizados com três repetições, 22 acessos e quatro plantas por parcela. Desenvolveram-se perceptrons de multicamadas utilizando 50 folhas por acesso, destinando-se 70% para treinamento, 15% para a validação cruzada (early-stop) e 15% para teste. Foram testadas 39 configurações de rede perceptron de multicamadas. As RNAs foram eficientes para estimar a área foliar da couve a partir do comprimento e largura do limbo foliar. A área foliar estimada pela RNA é indicada para a seleção de plantas por ser de fácil obtenção, ser um método não destrutivo, apresentar alta correlação fenotípica e genética com a área foliar observada e maior herdabilidade.

English

Forestry production is traditionally predicted using mathematical modelling, where whole-stand models are prominent for providing estimates of growth and production per unit area. However, there is a need to perform research that adopts innovative tools, such as artificial Intelligence techniques. The objective of this study was to train and evaluate the efficiency of Artificial Neural Networks (ANN) in the modeling process of growth and production of Whole-Stand Level, in “equineanean” forests of the Eucalyptus genus clones. For the training of the networks, the supervised method was adopted. There were 100 networks trained, of which, for each output variable, 5 networks were selected. The criteria used to verify the quality of the training and validation of the networks were: Student's test “T”, graphical analysis of the dispersion of residues, standard error of the relative estimate (Syx%), Pearson's correlation coefficient (r) between the observed and estimated values, ​​and aggregate difference (AD). The ANN selected in the training process, which estimated the volume variables, site index and future production, when validated did not differ statistically by the Test T and presented adequate statistical accuracy values ​​to the modeling process, with satisfactory correlation values ​​(r) between the observed and estimated values, low values ​​of standard error of the relative estimate (Syx%) and aggregate difference (AD). Artificial neural networks of the multilayer perceptron type are precise and efficient in the entire modeling process for whole-stand level of Eucalyptus plantations, therefore its use is recommended. Key words: Artificial intelligence, prognosis, site classification, Brazil.

Comparing the Efficiency of Artificial Neural Network and Gene Expression Programming in Predicting Coronary Artery Disease

Background: Angiography, as the gold standard for the diagnosis of coronary artery disease, has made an attempt to predict coronary artery disease by comparing the efficiency of gene expression programming, as a new data mining technique, and artificial neural network, as a conventional technique. Besides, the study went further to present the results of feature selection based on stepwise backward elimination, classification and regression tree. Methods: The subjects were assessed for nine coronary artery disease risk factors to develop a prediction model for the disease. They included 13,288 patients who were chosen to undergo angiography for the diagnosis of coronary artery disease; from this sample, 4059 subjects were free from the disease while 9169 were suffering from it. Modeling was carried out based on gene expression programming and artificial neural network techniques. The Delong’s test was then used to choose the final model based on the area under the Receiver Operating Characteristic (ROC) curve. Results: The model, developed based on artificial neural network, had AUC of 0.719, accuracy of 73.39%, sensitivity of 93.44% and specificity of 28.34%. On the other hand, the model, formulated based on gene expression programming, had AUC of 0.720, accuracy of 73.94%, sensitivity of 93.29% and specificity of 31.43%. Delong’s test showed no significant difference between the two models (p value=0/789). Then, feature selection method was used to choose a model with four risk factors and an accuracy rate of 73.26%. Conclusion: Comparison of the results showed no significant difference between the two modeling techniques. The gene expression programming model was very easy to present and interpret; it could also be easily converted to other programming languages; so, with these features in mind, the researchers preferred to choose this technique.

Monitoring of nanodiamonds in human urine using artificial neural networks

This paper presents the results of application of the elaborated methods for monitoring of nanodiamonds in human urine using fluorescence spectroscopy. High efficiency of artificial neural networks applied for recognition and estimation of the concentration of nanodiamonds in urine with a strong autofluorescence background is demonstrated. It was found that minimal concentration of nanodiamonds with strong fluorescence containing nitrogen‐vacancy (NV) centers can be detected in urine by fluorescence spectroscopy at a level of 3.06 × 10−4 g L−1; while use of artificial neural networks for detection of weakly fluorescent detonation nanodiamonds provides reasonably high accuracy of detection, not worse than 6.8 × 10−3 g L−1.

Recognition of culling reasons in Polish dairy cows using data mining methods

Cow longevity and reasons for culling are one of the most important research problems in the contemporary cattle breeding. Therefore, the analysis of the relationship between cow performance and involuntary disposal contributes to taking more informed decisions in herd management. The aim of the present study was to compare the efficiency of artificial neural networks (ANN) and boosted classification trees (BT) with that of linear discriminant analysis (LDA) and classification functions (CF) in recognizing culling reasons of dairy cows in Poland, based on the lifetime performance data, routinely monitored in a herd. The analyses carried out in the present study showed that the accurate recognition of different culling reasons based on predictors included in the above-mentioned models is, in general, impossible. Only BT had limited discrimination abilities, but the results obtained using this method were not much improved compared with ANN and LDA with CF. In order to predict precisely various culling reasons, more specific data are required. They could be obtained from the increasingly popular, technologically advanced, systems of real-time monitoring of animal health status (physical activity, rumination rate, etc.), dependent also on environmental conditions (e.g. temperature–humidity index).

Designing artificial neural network on thermal conductivity of Al2O3–water–EG (60–40 %) nanofluid using experimental data

The main purpose of this research was to investigate the efficiency of artificial neural networks in modeling thermal conductivity data of water–EG (40–60 %) nanofluid with aluminum oxide nanoparticles (with average diameter of 36 nm). The measured data as modeling input data are in six volume fractions from 0 to 1.5 % and different temperatures from 20 to 60 °C. In order to optimize the network, different numbers of neurons with different transfer functions have been tested and after preprocessing and normalizing the data, the optimum network structure with one hidden layer and six neurons was obtained. This structure simulated the experimental data with very high precision. The measured thermal conductivity was compared with the two models that calculated thermal conductivity for mixtures. The results indicated that Hamilton–Crosser and Lu–Lin models failed in estimating the thermal conductivity of Alumina–water–EG nanofluid in different temperatures and concentration. Finally, a new correlation was presented based on experimental data with regression coefficient of 0.9974.

On the efficiency of artificial neural networks for plastic analysis of planar frames in comparison with genetic algorithms and ant colony systems

The investigation of plastic behavior and determining the collapse load factors are the important ingredients of every kinematical method that is employed for plastic analysis and design of frames. The determination of collapse load factors depends on many effective parameters such as the length of bays, height of stories, types of loads and plastic moments of individual members. As the number of bays and stories increases, the parameters that have to be considered make the analysis a complex and tedious task. In such a situation, the role of algorithms that can help to compute an approximate collapse load factor in a reasonable time span becomes more and more crucial. Due to their interesting properties, heuristic algorithms are good candidates for this purpose. They have found many applications in computing the collapse load factors of low-rise frames. In this work, artificial neural networks, genetic algorithms and ant colony systems are used to obtain the collapse load factors of two-dimensional frames. The latter two algorithms have already been employed in the analysis of frames, and hence, they provide a good basis for comparing the results of a newly developed algorithm. The structure of genetic algorithm, in the form presented here, is the same as previous works; however, some minor amendments have been applied to ant colony systems. The performance of each algorithm is studied through numerical examples. The focus is mainly on the behavior of artificial neural networks in the determination of collapse load factors of two-dimensional frames compared with other two algorithms. The investigation of results shows that a careful selection of the structure of artificial neural networks can lead to an efficient algorithm that predicts the load factors with higher accuracy. The structure should be selected with the aim to reduce the error of the network for a given frame. Such an algorithm is especially useful in designing and analyzing frames whose geometry is known a priori.

Shear strength estimation of RC deep beams using the ANN and strut-and-tie approaches

Reinforced concrete (RC) deep beams are structural members that predominantly fail in shear. Therefore, determining the shear strength of these types of beams is very important. The strut-and-tie method is commonly used to design deep beams, and this method has been adopted in many building codes (ACI318-14, Eurocode 2-2004, CSA A23.3-2004). In this study, the efficiency of artificial neural networks (ANNs) in predicting the shear strength of RC deep beams is investigated as a different approach to the strut-and-tie method. An ANN model was developed using experimental data for 214 normal and high-strength concrete deep beams from an existing literature database. Seven different input parameters affecting the shear strength of the RC deep beams were selected to create the ANN structure. Each parameter was arranged as an input vector and a corresponding output vector that includes the shear strength of the RC deep beam. The ANN model was trained and tested using a multi-layered back-propagation method. The most convenient ANN algorithm was determined as trainGDX. Additionally, the results in the existing literature and the accuracy of the strut-and-tie model in ACI318-14 in predicting the shear strength of the RC deep beams were investigated using the same test data. The study shows that the ANN model provides acceptable predictions of the ultimate shear strength of RC deep beams (maximum <TEX>$R^2{\approx}0.97$</TEX>). Additionally, the ANN model is shown to provide more accurate predictions of the shear capacity than all the other computed methods in this study. The ACI318-14-STM method was very conservative, as expected. Moreover, the study shows that the proposed ANN model predicts the shear strengths of RC deep beams better than does the strut-and-tie model approaches.

The Efficiency of Artificial Neural Networks for Forecasting in the Presence of Autocorrelated Disturbances

&lt;p&gt;We compare three forecasting methods, Artificial Neural Networks (ANNs), Autoregressive Integrated Moving Average (ARIMA) and Regression models. Using computer simulations, the major finding reveals that in the presence of autocorrelated errors ANNs perform favorably compared to ARIMA and regression for nonlinear models. The model accuracy for ANN is evaluated by comparing the simulated forecast results with the real data for unemployment in Palestine which were found to be in excellent agreement.&lt;/p&gt;

Use of interactive multisensor snow and ice mapping system snow cover maps (IMS) and artificial neural networks for simulating river discharges in Eastern Turkey

Basins located in Eastern Turkey are largely fed by snowmelt runoff during spring and early summer seasons. This study investigates the efficiency of artificial neural networks (ANNs) in snowmelt runoff generation. Although ANNs have been used for streamflow simulating/forecasting in the last two decades, using satellite-based snow-covered area (SCA) maps and meteorological observations as inputs to ANN provides a novel basis for estimating streamflow. The proposed methodology is implemented over Upper Euphrates River Basin in Eastern Turkey. SCA data was acquired from Interactive Multisensor Snow and Ice Mapping System (IMS) for an 8-year period from February 2004 to September 2011. Meteorological observations including daily cumulative precipitation and daily average air temperatures were obtained from Turkish State Meteorological Services. The simulation results are promising with coefficient of correlation varying from 0.67 to 0.98 among proposed models. Past days discharge was found to substantially improve the forecast accuracy. The paper presents the expected basin discharge for 2011 water year based on meteorological observations and SCA input.

Prediction of Land Surface Temperature (LST) Changes within Ikom City in Nigeria Using Artificial Neural Network (ANN)

Land Surface Temperature (LST) is one of the factors associated to urban heat rise and micro climatic warming within a city. Researches relating to the development new technologies or the improvement on the existing ones are very important in urban climate studies. This paper expounds our study on the simulation and prediction of specific future time LST quantitative trend in Ikom city of Nigeria using Feed Forward Back Propagation Artificial Neural Network technology. This study was based on time series ANN model that takes a sequence of past LST values, understand the pattern of change within the dataset and further predict or future time values. Similar studies have been carried out in this manner from our literature review but none used earth observation time series satellite data of a coarse resolution epoch interval for LST time series prediction using ANN. The novelty of this study centers on the attempt to predict some specific future time LST values city-wide using ANN from past LST values derived from earth observation remote sensing imagery (Landsat 7 ETM). The results derived from this study reaffirm the efficiency of ANN (part of deep learning technologies) in learning, understanding and making accurate predictions from a non-linear chaotic real world complex dataset.

Dynamic modeling and optimization of cyanobacterial C-phycocyanin production process by artificial neural network

Artificial neural networks have been widely applied in bioprocess simulation and control due to their advantageous properties. However, their feasibility in long-term photo-production process modeling and prediction as well as their efficiency on process optimization have not been well studied so far. In the current study, an artificial neural network was constructed to simulate a 15-day fed-batch process for cyanobacterial C-phycocyanin production, which to the best of our knowledge has never been conducted. To guarantee the accuracy of artificial neural network, two strategies were implemented. The first strategy is to generate artificial data sets by adding random noise to the original data set, and the second is to choose the change of state variables as training data output. In addition, the first strategy showed the distinctive advantage of reducing the experimental effort in generating training data. By comparing with current experimental results, it is concluded that both strategies give the network great modeling and predictive power to estimate the entire fed-batch process performance, even when few original experimental data are supplied. Furthermore, by optimizing the operating conditions of a 12-day fed-batch process, a significant increase of 85.6% on C-phycocyanin production was achieved compared to previous work, which suggests the high efficiency of artificial neural network on process optimization.

Fiber nonlinearity-induced penalty reduction in CO-OFDM by ANN-based nonlinear equalization.

We experimentally demonstrate ∼2 dB quality (Q)-factor enhancement in terms of fiber nonlinearity compensation of 40 Gb/s 16 quadrature amplitude modulation coherent optical orthogonal frequency-division multiplexing at 2000km, using a nonlinear equalizer (NLE) based on artificial neural networks (ANN). Nonlinearity alleviation depends on escalation of the ANN training overhead and the signal bit rate, reporting ∼4 dBQ-factor enhancement at 70 Gb/s, whereas a reduction of the number of ANN neurons annihilates the NLE performance. An enhanced performance by up to ∼2 dB in Q-factor compared to the inverse Volterra-series transfer function NLE leads to a breakthrough in the efficiency of ANN.

Efficiency of artificial neural networks in map of total electron content over Iran

Maps of the total electron content (TEC) of the ionosphere can be reconstructed using data extracted from global positioning system (GPS) signals. For historic and other sparse data sets, the reconstruction of TEC images is often performed using multivariate interpolation techniques. In this paper, a quantitative comparison of the ability of artificial neural networks (ANN), polynomial fitting and kriging interpolation was carried out in order to model the spatial variations of TEC using GPS data over Iran. These methods are suitable for handling multi-scale phenomena and unevenly distributed data. The observations collected at 25 GPS stations from Iranian permanent GPS network (uniformly spread all over Iran with sampling rate of 30-seconds). Dual frequency carrier phase and code GPS observations were used. A smoothed TEC approach was used for absolute TEC recovery. Evaluation of the methods has been applied with single GPS station in Tehran equipped with ionosonde instrument. The minimum relative error for ANN, polynomial and kriging are 4.37, 6.35, 9.13 % and the maximum relative error are 8.61, 29.06, and 20.14 % respectively. Also root mean square error (RMSE) of 3.7 TECU is computed for ANN method which is less than RMSE of other mentioned methods. The results show that ANN method has higher accuracy and compiles speed than kriging and polynomial. As well as, it is found that polynomial and kriging methods required many computational points in adjustment step.

Application of artificial neural networks in indirect selection: a case study on the breeding of lettuce

The efficiency of artificial neural networks (ANN) to model complex problems may enable the prediction of characteristics that are hard to measure, providing better results than the traditional indirect selection. Thus, this study aimed to investigate the potential of using artificial neural networks (ANN) for indirect selection against early flowering in lettuce, identify the influence of genotype by environment interaction in this strategy and compare your results with the traditional indirect selection. The number of days to anthesis were used as the desired output and the information of six characteristics (fresh weight of shoots, mass of marketable fresh matter of shoots, commercial dry matter of shoots, average diameter of the head, head circumference and leaf number) as input file for the training of the ANN-MLP (Perceptron Multi-Layer). The use of ANN has great potential adjustment for indirect selection for genetic improvement of lettuce against early flowering. The selection based on the predicted values by network provided estimates of gain selection largest that traditional indirect selection. The ANN trained with data from an experiment have low power extrapolation to another experiment, due to effect of interaction genotype by environment. The ANNs trained simultaneously with data from different experiments presented greater predictive power and extrapolation.

Comparative Efficiency of Artificial Neural Network to Predict Spring-Back Value in U-Bending Process of High Strength Steel

High strength steel (HSS) was used as raw material in the automotive industry, which required lightweight and high strength, such as chassis and etc. However, the strength and hardness of the steel are relatively high, leading to the low permeability and large spring-back occurring after forming operation. As a result, the work piece is not shaped desire. This research proposes neural network for predicts the spring back values, in U-bending process, that the materials were differences in mechanical properties, such as SPFH590 (JIS) and SPEC980Y (JIS). In the experiment, the input factors for predictable data consists as the punch radius (RP), die radius (Rd), clearance (Cl) and counter punch force (Fc). After that, the input data were analyzed relation with spring-back values by the Pearson Correlation of One-tailed. Next, It was selected by Leave-one-out and k-fold Cross validation (K-fold and LOOCV), to improve efficiency of the prediction process. Moreover, the result was a measurable performance with Root Mean Square Error (RMSE) technique, equal to 0.788 and 2.10 respectively. In the final analysis, the neural network is effective to predict the Spring-back values of SPFH590 (JIS) rather than SPEC 980Y (JIS) in U-bending process.

Development of a novel flyrock distance prediction model using BPNN for providing blasting operation safety

One of the threatening safety problems in mines is flyrock distance range through blasting operation. Inaccurate evaluation of flyrock can cause fatal and nonfatal accidents. The presented results in this paper verify efficiency of artificial neural network in prediction of flyrock considering all influencing parameters such as: hole diameter, height, subdrilling, number of holes, spacing, burden, ANFO amount, dynamite weight, stemming, powder factor, specific drilling, and delay time. In this research, optimum structure of network was determined by studying different transfer functions and number of the neurons using a programming code. In this case, optimum structure configuration is logsig transfer functions for the two hidden layers and tansig or logsig one for output, and there are eight neurons in each hidden layers. By calculating strength of relationship between flyrock and all influencing parameters using cosine amplitude method (CAM), the powder factor is defined as most effective parameter on the flyrock.

Artificial neural networks reveal efficiency in genetic value prediction.

The objective of this study was to evaluate the efficiency of artificial neural networks (ANNs) for predicting genetic value in experiments carried out in randomized blocks. Sixteen scenarios were simulated with different values of heritability (10, 20, 30, and 40%), coefficient of variation (5 and 10%), and the number of genotypes per block (150 and 200 for validation, and 5000 for neural network training). One hundred validation populations were used in each scenario. Accuracy of ANNs was evaluated by comparing the correlation of network value with genetic value, and of phenotypic value with genetic value. Neural networks were efficient in predicting genetic value with a 0.64 to 10.3% gain compared to the phenotypic value, regardless the simulated population size, heritability, or coefficient of variation. Thus, the artificial neural network is a promising technique for predicting genetic value in balanced experiments.