Performance Of Artificial Neural Network Model Research Articles

Artificial neural network predictions of water levels in a Gulf of Mexico shallow embayment

Tide tables are the method of choice for water level predictions in most coastal regions. However, for many locations along the coast of the Gulf of Mexico, tide tables do not meet United States National Ocean Service (NOS) standards. Wind forcing has been recognized as the main variable not included. The performance of the tide tables is particularly poor in shallow embayments. Recent research has shown that Artificial Neural Network (ANN) models including input variables such as previous water levels, tidal forecasts, wind speed, wind direction, wind forecasts and barometric pressure can greatly improve over the tide charts for locations including open coast and deep embayments. In this paper, the ANN modeling technique is applied to a shallow embayment, the station of Rockport, located near Corpus Christi, Texas. The ANN model performance is compared against the NOS tide charts and the persistence model for the years 1997 to 2001. The performance is assessed using NOS criteria including Central Frequency (CF of 15 cm), Maximum Duration of Positive Outliers (MDPO), and Maximum Duration of Negative Outliers (MDNO). Over the study period, the performances of the three models (tide table, persistence, ANN) are respectively CF’s of 85%, 95.8% and 96.9%, MDPOs of 16, 14 and 5.9 hours, and MDNOs of 72.8 hours, 0.6 and 9.5 hours.

Artificial neural networks for corn and soybean yield prediction

The Maryland Water Quality Improvement Act of 1998 requires mandatory nutrient management planning on all agricultural land in Maryland. Nutrient management specialists need simple and accurate estimation techniques to relate crop yields and nutrient utilization in the planning process. The objectives of this study were to: (1) investigate if artificial neural network (ANN) models could effectively predict Maryland corn and soybean yields for typical climatic conditions; (2) compare the prediction capabilities of models at state, regional, and local levels; (3) evaluate ANN model performance relative to variations of developmental parameters; and (4) compare the effectiveness of multiple linear regression models to ANN models. Models were developed using historical yield data at multiple locations throughout Maryland. Field-specific rainfall data and the USDA-NRCS Soil Rating for Plant Growth (SRPG) values were used for each location. SRPG and weekly rainfall means were necessary for effective corn and soybean yield predictions. Adjusting ANN parameters such as learning rate and number of hidden nodes affected the accuracy of crop yield predictions. Optimal learning rates fell between 0.77 and 0.90. Smaller data sets required fewer hidden nodes and lower learning rates in model optimization. ANN models consistently produced more accurate yield predictions than regression models. ANN corn yield models for Maryland resulted in r 2 and RMSEs of 0.77 and 1036 versus 0.42 and 1356 for linear regression, respectively. ANN soybean yield models for Maryland resulted in r 2 and RMSEs of 0.81 and 214 versus 0.46 and 312 for linear regression, respectively. Although more time consuming to develop than linear regression models, ANN models proved to be a superior methodology for accurately predicting corn and soybean yields under typical Maryland climatic conditions.

Modelling beta transus temperature of titanium alloys using artificial neural network

An artificial neural network (ANN) model is developed to simulate the non-linear relationship between the beta transus (βtr) temperature of titanium alloys and the alloy chemistry. The input parameters to the model consist of the concentration of nine elements, i.e. Al, Cr, Fe, Mo, Sn, Si, V, Zr and O, whereas the model output is the βtr temperature. Good performance of the ANN model was achieved. The interactions between the alloying elements were estimated based on the obtained ANN model. The results showed good agreement with experimental data. The influence of the database scale on ANN model performance was also discussed. Estimation of βtr temperature through thermodynamic calculation was carried out as a comparison.

Prediction of the affinity of the newly synthesised azapirone derivatives for 5-HT1A receptors based on artificial neural network analysis of chromatographic retention data and calculation chemistry parameters.

The performance of artificial neural network (ANN) in predicting the affinity of a series of 65 new azapirone derivatives for rat brain serotonin 5-HT1A receptors based on high-performance liquid chromatography (HPLC) retention data and on non-empirical structural parameters of the compounds' was studied. Affinity of the agents for rat brain 5-HT1A receptors were assessed in vitro and expressed as inhibitor constant values, Ki. The retention parameters determined in 14 HPLC systems along with compounds' structural descriptors from calculation chemistry were considered in ANN analysis. Supervised method of ANN learning with back-propagation strategy was used in ANN calculations. Two models of ANN of similar architecture were designed: the first one for the data based on chromatographic retention data and the second based on structural parameters of the agents. Each ANN model was trained with the data of training set. It was next used to classify the agents from the testing set into two groups: active (Ki < 50 nM) and inactive compounds (Ki > 50 nM). A high prediction performance of both ANN models considered as regards the affinity of new azapirone derivatives for the serotonin 5-HT1A receptors was demonstrated. However, the percent of correctly classified compounds was higher in the case of the ANN processing of the non-empirical structural descriptors of azapirone derivatives. Since the ANN analysis of the retention data and of the structural parameters originating from calculation chemistry allows to segregate drug candidates according to their pharmacological properties that, in consequence, may be of help to limit the number of biological assays in the search for new drugs.

Data Division for Developing Neural Networks Applied to Geotechnical Engineering

In recent years, artificial neural networks (ANNs) have been applied to many geotechnical engineering problems with some degree of success. In the majority of these applications, data division is carried out on an arbitrary basis. However, the way the data are divided can have a significant effect on model performance. In this paper, the issue of data division and its impact on ANN model performance is investigated for a case study of predicting the settlement of shallow foundations on granular soils. Four data division methods are investigated: (1) random data division; (2) data division to ensure statistical consistency of the subsets needed for ANN model development; (3) data division using self-organizing maps (SOMs); and (4) a new data division method using fuzzy clustering. The results indicate that the statistical properties of the data in the training, testing, and validation sets need to be taken into account to ensure that optimal model performance is achieved. It is also apparent from the results that the SOM and fuzzy clustering methods are suitable approaches for data division.

Artificial Neural Networks for Adolescent Marijuana Use and Clinical Features of Marijuana Dependence

This article compares the performance of multiple logistic regression (MLR) with feed-forward, artificial neural network (ANN) models for the assessment of adolescent marijuana use and clinical features of dependence based on self-evaluation from recent National Household Surveys on Drug Abuse (NHSDA). The effect of training and testing the neural networks with randomly selected data was compared to data selected as a function of survey year. The technical aim of the study was to account for adolescent marijuana use and features of marijuana dependence based on experiences with alcohol and tobacco. Similarities observed in MLR and ANN model performance may indicate no major complex or nonlinear relationships in cross-sectional epidemiological data selected to model adolescent drug use and dependence in this specific application. We concluded that ANNs should be further studied in future longitudinal research, perhaps with modeling of recursive networks, allowing feedback from drug dependence to levels of marijuana use. The ANN models also have the potential to model drug use and dependence based on input parameters with no obvious direct link to drug involvement—e.g., polymorphisms associated with “openness to experience” or other personality traits hypothesized to function as distal antecedents, and could thus be implemented to identify higher risk youths using assessments indirectly related or nonlinearly associated to adolescent drug use and dependence but less sensitive to survey-related response tendencies.

Modelling and forecasting of hydrological variables using artificial neural networks: the Kafue River sub-basin

A simple reservoir routing scheme is applied to the available hydrometeorlogical data from the Kafue River sub-drainage basin in Zambia to derive flow contributions from the ungauged parts of the basin. The derived flow series and the time series of historic flow measured at the Kafue Hook Bridge (KHB) are separately modelled using artificial neural networks (ANNs). For each of these two flow series, relevant input variables are determined with the help of input—output intercorrelations, where inputs are given to a host of three-layer feedforward back-propagation (FF-BP) ANNs to predict the current, derived flow or KHB flow. A couple of ANN models selected on the basis of defined criteria are then used to forecast the flows at m time steps ahead. To evaluate the forecasting performance of the best ANN models, comparison with best autoregressive moving average models with exogenous inputs, ARMAX, is made. In both cases the ANNs give more robust forecasts over long terms than the ARMAX models, thereby making ANNs a viable alternative in flow forecasting.

Methods to improve prediction performance of ANN models

Artificial neural network (ANN) is a powerful tool and applied successfully in numerous fields. But there are still two limitations on its use. One is over-training, which occurs when the capacity of the ANN for training is too great because it is allowed too many training iterations. The other is that ANNs are not effective for extrapolation, which is sometimes very important because the existing data used to train an ANN do not necessarily cover the entire range. The two limitations degrade seriously the prediction performance of ANN models. In this paper, two practices are introduced to alleviate or overcome the negative effect of the limitations. Demonstrations based on these practices indicate that they are general and useful practices and can improve greatly the prediction performance of the resulting ANN models to make them really suitable for engineering applications.

Using neural networks for calibration of time-domain reflectometry measurements

Time-domain reflectometry (TDR) is an electromagnetic technique for measurements of water and solute transport in soils. The relationship between the TDR-measured dielectric constant (Ka ) and bulk soil electrical conductivity ([sgrave]a) to water content (θW) and solute concentration is difficult to describe physically due to the complex dielectric response of wet soil. This has led to the development of mostly empirical calibration models. In the present study, artificial neural networks (ANNs) are utilized for calculations of θw and soil solution electrical conductivity ([sgrave]w) from TDR-measured Ka and [sgrave]a in sand. The ANN model performance is compared to other existing models. The results show that the ANN performs consistently better than all other models, suggesting the suitability of ANNs for accurate TDR calibrations.

Neural network based modelling of environmental variables: A systematic approach

Feedforward artificial neural networks (ANNs) that are trained with the back-propagation algorithm are a useful tool for modelling environmental systems. They have already been successfully used to model salinity, nutrient concentrations, air pollution, and algal growth. These successes, coupled with their suitability for modelling complex systems, have resulted in an increase in their popularity and their application in an ever increasing number of areas. They are generally treated as black box models that are able to capture underlying relationships when presented with input and output data. In many instances, little consideration is given to potential input data and the internal workings of ANNs. This can result in inferior model performance and an inability to accurately compare the performance of different ANN models. Back-propagation networks employ a modelling philosophy that is similar to that of statistical methods in the sense that unknown model parameters (i.e., connection weights) are adjusted in order to obtain the best match between a historical set of model inputs and corresponding outputs. Consequently, the principles that are considered good practice in the development of statistical models should be considered. In this paper, a systematic approach to the development of ANN based forecasting models is presented, which is intended to act as a guide for potential and current users of feedforward ANNs that are trained with the back-propagation algorithm. Issues that need to be considered in the model development phase are discussed and ways of addressing them presented. The major areas covered include data transformation, the determination of appropriate model inputs, the determination of an appropriate network geometry, the optimisation of connection weights, and validation of model performance.

Hydrological modelling using artificial neural networks

This review considers the application of artificial neural networks (ANNs) to rainfall-runoff modelling and flood forecasting. This is an emerging field of research, characterized by a wide variety of techniques, a diversity of geographical contexts, a general absence of intermodel comparisons, and inconsistent reporting of model skill. This article begins by outlining the basic principles of ANN modelling, common network architectures and training algorithms. The discussion then addresses related themes of the division and preprocessing of data for model calibration/validation; data standardization techniques; and methods of evaluating ANN model performance. A literature survey underlines the need for clear guidance in current modelling practice, as well as the comparison of ANN methods with more conventional statistical models. Accordingly, a template is proposed in order to assist the construction of future ANN rainfall-runoff models. Finally, it is suggested that research might focus on the extraction of hydrological ‘rules’ from ANN weights, and on the development of standard performance measures that penalize unnecessary model complexity.

Rainfall Disaggregation Using Artificial Neural Networks

A precipitation time series is often a necessary input for the analysis and design of hydrologic and hydraulic systems. The precipitation records employed for these purposes can be either measured observations or generated by stochastic simulation. One common problem with recorded and generated precipitation data is that often it is not in small enough time increments for use in engineering applications (e.g., continuous hydrologic simulation). To solve this problem, rainfall amounts can be disaggregated into shorter time increments. This paper evaluates the use of artificial neural networks (ANNs) for the disaggregation of hourly rainfall data into subhourly time increments. Two different ANN models are introduced and evaluated in this paper. A back-propagation/steepest-descent algorithm trains one model and the other model uses the idea of self-organization in a competitive learning ANN. The results indicate that the performance of both ANN models are: (1) comparable to other disaggregation schemes in terms of predicting the overall disaggregated rainfall hyetograph; and (2) improvements over other disaggregation models in the prediction of the maximum incremental rainfall intensity (depth) within a storm hyetograph. Based on these results, the use of ANN models can be recommended as a viable alternative for hourly rainfall disaggregation.

The Prediction of Earnings Using Financial Statement Information: Empirical Evidence With Logit Models and Artificial Neural Networks

In the past three decades, earnings have been one of the most researched variables in accounting. Empirical research provided substantial evidence on its usefulness in the capital markets but evidence in predicting earnings has been limited, yielding inconclusive results. The purpose of this study is to validate and extend prior research in predicting earnings by examining aggregate and industry-specific data. A sample of 10,509 firm-year observations included in the Compustat database for the period 1982–91 is used in the study. The stepwise logistic regression results of the present study indicated that nine earnings and non-earnings variables can be used to predict earnings. These predictor variables are not identical to those reported in prior studies. These results are also extended to the manufacturing industry. Two new variables are identified to be significant in this industry. Moreover, an Artificial Neural Network (ANN) approach is employed to complement the logistic regression results. The ANN model's performance is at least as high as the logistic regression model's predictive ability. © 1996 Wiley Periodicals, Inc.