Deep Artificial Neural Networks Research Articles

Electricity peak shaving for commercial buildings using machine learning and vehicle to building (V2B) system

Reducing electricity peak demand is essential to maintain the balance between supply and demand side in electricity power markets, as well as reduce utility costs and environmental impacts. With growth in adoption of Electric Vehicles (EVs), there is an emerging opportunity to balance electrical power demand of buildings by storing electricity in EVs during low demand periods and discharging electricity into buildings during peak demand periods. Due to uncertainty in time and magnitude of peak demand, decision makers are always faced with a challenging task to identify optimal schedules for charging and discharging EVs to minimize peak electricity demand. This paper presents the development of a novel system that is capable of predicting day-ahead building electricity demand profile and identifying optimum schedule of charging and discharging EVs to minimize electricity peak demand. The system is designed to comply with planned EV trip schedules and minimum state of charge (SOC). The system consists of (1) machine Learning (ML) model to predict electrical power demand, and (2) demand management optimization model to identify optimal schedule for charging and discharging EVs. Four methods are explored to develop the ML model, including histogram-based gradient boosting, random forest, deep artificial neural network (DNN), and long short-term memory (LSTM). A case study of multi-tenant commercial building is analyzed to evaluate the performance of the system and demonstrate its new capabilities. The results of the case study shows that LSTM has the best performance in terms of mean absolute error, root mean square error, and mean absolute percentage error with average values of 7.44, 17.78, and 20.08 %, respectively. Five scenarios for shaving peak electricity demand, including combinations of two electric vehicles, a stationary battery, and a PV system are investigated. Scenarios including the stationary battery and the PV system are considered to evaluate the full potential of peak demand reduction in the case study building. The results of the demand management optimization model show up to 36 % reduction in peak demand using two EVs, one stationary battery, and PV system of 40 kW capacity. The key contributions that this study adds to existing knowledge are: (1) developing machine learning models to predict day-ahead electricity demand in 15-minute intervals, (2) integrating machine learning and optimization algorithms in identifying EV charging and discharging schedules to minimize utility cost by shaving peak energy demand, and (3) considering planned EV trips and minimum SOC requirements in identifying optimal charging and discharging of EVS to shave peak energy demand in buildings. The implementation of this system provides practical solutions for managing electricity demand in commercial buildings using EVs. By reducing energy consumption and promoting the innovative use of EVs, this system offers a sustainable approach to managing electricity demand in commercial buildings.

Performance prediction of the PEMFCs based on gate recurrent unit network optimized by improved version of prairie dog optimization algorithm

In this article, the black box dynamic model is presented for forecasting the performance of the PEM (Proton-exchange membrane) fuel cell (FC). An optimized deep artificial neural network has been used to build the experimental nonlinear model of the polymer membrane FC series that functions with hydrogen and oxygen. This research investigates predictability for a gate recurrent unit (GRU) optimized by a modified Prairie Dog Optimizer in PEMFCs. The results obtained have been validated by applying a case study and then a comparison is conducted among the outcomes of the offered technique and 2 other published methods: modified relevance vector machine and Lattice Gated Recurrent Unit (LGRU). The voltage clearly changes significantly, as demonstrated by simulations, even though the FC is handled with a low starting temperature and current. Also, the voltage point distribution has become more concentrated when the current and temperature are high. In both the training and prediction phases, the MAPE is reduced to approximately 0.0043 and 0.0047, respectively, showing that the proposed GRU technique produces superior prediction results when the operational settings approach the optimum operating conditions. According to simulations, the proposed IPDO/GRU with a 0.004 root mean square has the least error, followed by the mRVM and GRU with 0.009 and 0.010 root mean square values. The outcomes show that using the offered procedure does provide the finest verification of the empirical data.

Artificial Deep Neural Network for Sensorless Pump Flow and Hemodynamics Estimation During Continuous-Flow Mechanical Circulatory Support.

The objective of this study was to compare the estimates of pump flow and systemic vascular resistance (SVR) derived from a mathematical regression model to those from an artificial deep neural network (ADNN). Hemodynamic and pump-related data were generated using both the Cleveland Clinic continuous-flow total artificial heart (CFTAH) and pediatric CFTAH on a mock circulatory loop. An ADNN was trained with generated data, and a mathematical regression model was also generated using the same data. Finally, the absolute error for the actual measured data and each set of estimated data were compared. A strong correlation was observed between the measured flow and the estimated flow using either method (mathematical, R = 0.97, p < 0.01; ADNN, R = 0.99, p < 0.01). The absolute error was smaller in the ADNN estimation (mathematical, 0.3 L/min; ADNN 0.12 L/min; p < 0.01). Furthermore, strong correlation was observed between measured and estimated SVR (mathematical, R = 0.97, p < 0.01; ADNN, R = 0.99, p < 0.01). The absolute error for ADNN estimation was also smaller than that of the mathematical estimation (mathematical, 463 dynes·sec·cm -5 ; ADNN, 123 dynes·sec·cm -5 , p < 0.01). Therefore, in this study, ADNN estimation was more accurate than mathematical regression estimation. http://links.lww.com/ASAIO/A991.

Color Image Encryption using Chaotic Algorithm and 2D Sin-Cos Henon Map for High Security

In every form of electronic communication, data security must be an absolute top priority. As the prevalence of Internet and other forms of electronic communication continues to expand, so too does the need for visual content. There are numerous options for protecting transmitted data. It's important that the transmission of hidden messages in images remain unnoticed to avoid raising any red flags. In this paper, we propose a new deep learning-based image encryption algorithm for safe image retrieval. The proposed algorithm employs a deep artificial neural network model to extract features via sample training, allowing for more secure image network transmission. The algorithm is incorporated into a deep learning-based image retrieval process with Convolution Neural Networks(CNN), improving the efficiency of retrieval while also guaranteeing the security of ciphertext images. Experiments conducted on five different datasets demonstrate that the proposed algorithm vastly improves retrieval efficiency and strengthens data security. Also hypothesised a 2D Sin-Cos-Henon (2D-SCH)-based encryption algorithm for highly secure colour images. We demonstrate that this algorithm is secure against a variety of attacks and that it can encrypt all three colour channels of an image simultaneously.

Deep learning delay coordinate dynamics for chaotic attractors from partial observable data.

A common problem in time-series analysis is to predict dynamics with only scalar or partial observations of the underlying dynamical system. For data on a smooth compact manifold, Takens' theorem proves a time-delayed embedding of the partial state is diffeomorphic to the attractor, although for chaotic and highly nonlinear systems, learning these delay coordinate mappings is challenging. We utilize deep artificial neural networks (ANNs) to learn discrete time maps and continuous time flows of the partial state. Given training data for the full state, we also learn a reconstruction map. Thus, predictions of a time series can be made from the current state and several previous observations with embedding parameters determined from time-series analysis. The state space for time evolution is of comparable dimension to reduced order manifold models. These are advantages over recurrent neural network models, which require a high-dimensional internal state or additional memory terms and hyperparameters. We demonstrate the capacity of deep ANNs to predict chaotic behavior from a scalar observation on a manifold of dimension three via the Lorenz system. We also consider multivariate observations on the Kuramoto-Sivashinsky equation, where the observation dimension required for accurately reproducing dynamics increases with the manifold dimension via the spatial extent of the system.

Equivalent processing of facial expression and identity by macaque visual system and task-optimized neural network

Both the primate visual system and artificial deep neural network (DNN) models show an extraordinary ability to simultaneously classify facial expression and identity. However, the neural computations underlying the two systems are unclear. Here, we developed a multi-task DNN model that optimally classified both monkey facial expressions and identities. By comparing the fMRI neural representations of the macaque visual cortex with the best-performing DNN model, we found that both systems: (1) share initial stages for processing low-level face features which segregate into separate branches at later stages for processing facial expression and identity respectively, and (2) gain more specificity for the processing of either facial expression or identity as one progresses along each branch towards higher stages. Correspondence analysis between the DNN and monkey visual areas revealed that the amygdala and anterior fundus face patch (AF) matched well with later layers of the DNN's facial expression branch, while the anterior medial face patch (AM) matched well with later layers of the DNN's facial identity branch. Our results highlight the anatomical and functional similarities between macaque visual system and DNN model, suggesting a common mechanism between the two systems.

Deep Artificial Neural Network based Blind Color Image Watermarking in YCbCr Color Domain using statistical features

Deep Artificial Neural Network based Blind Color Image Watermarking in YCbCr Color Domain using statistical features

Deep Learning to Detect Pancreatic Cystic Lesions on Abdominal Computed Tomography Scans: Development and Validation Study.

Pancreatic cystic lesions (PCLs) are frequent and underreported incidental findings on computed tomography (CT) scans and can evolve to pancreatic cancer-the most lethal cancer, with less than 5 months of life expectancy. The aim of this study was to develop and validate an artificial deep neural network (attention gate U-Net, also named "AGNet") for automated detection of PCLs. This kind of technology can help radiologists to cope with an increasing demand of cross-sectional imaging tests and increase the number of PCLs incidentally detected, thus increasing the early detection of pancreatic cancer. We adapted and evaluated an algorithm based on an attention gate U-Net architecture for automated detection of PCL on CTs. A total of 335 abdominal CTs with PCLs and control cases were manually segmented in 3D by 2 radiologists with over 10 years of experience in consensus with a board-certified radiologist specialized in abdominal radiology. This information was used to train a neural network for segmentation followed by a postprocessing pipeline that filtered the results of the network and applied some physical constraints, such as the expected position of the pancreas, to minimize the number of false positives. Of 335 studies included in this study, 297 had a PCL, including serous cystadenoma, intraductal pseudopapillary mucinous neoplasia, mucinous cystic neoplasm, and pseudocysts . The Shannon Index of the chosen data set was 0.991 with an evenness of 0.902. The mean sensitivity obtained in the detection of these lesions was 93.1% (SD 0.1%), and the specificity was 81.8% (SD 0.1%). This study shows a good performance of an automated artificial deep neural network in the detection of PCL on both noncontrast- and contrast-enhanced abdominal CT scans.

Forecasting road traffic accident using deep artificial neural network approach in case of Oromia Special Zone

Millions of people are dying, and billions of properties are damaged by road traffic accidents each year worldwide. In the case of our country Ethiopia, the effect of traffic accidents is even more by causing injuries, death, and property damage. Forecasting road traffic accident and predicting the severity of road traffic accident contributes a role indirectly in reducing road traffic accidents. This study deals with forecasting the number of accident and prediction of the severity of an accident in the Oromia Special Zone using deep artificial neural network models. Around 6170 road traffic accidents data are collected from Oromia Police Commission Excel data and Oromia Special Zone Traffic Police Department; the dataset consists of accidents in the Special Zone of Oromia Districts (Woredas) from 2005 to 2012 with 15 accidents attributes. 5928 or (80%) of the dataset was used for the training model, and 1482 or (20%) of the dataset was used for the testing model. This study proposed six different neural network architectures such as backpropagation neural network (BPNN), feedforward neural network (FFNN), multilayer perceptron neural network (MLPNN), recurrent neural networks (RNN), radial basis function neural network (RBFNN) and long short-term memory (LSTM) models for accident severity prediction and the LSTM model for a time serious forecasting of number accidents within specified years. The models will take input data, classify accidents, and predict the severity of an accident. Accident predictor GUI has been created using Python Tkinter library for easy accident severity prediction. According to the model performance results, RNN model showed the best prediction accuracy of 97.18%, whereas MLP, LTSM, RBFNN, FFNN, and BPNN models showed the accuracy of 97.13%, 91.00%, 87.00%, 80.56%, and 77.26%, respectively. LTSM model forecasted accident for three years which is 3555 where the actual accident number is 3561. The prediction and forecast result obtained from the model will be helpful in planning and management of road traffic accidents.

Identifying regulation with adversarial surrogates

Homeostasis, the ability to maintain a relatively constant internal environment in the face of perturbations, is a hallmark of biological systems. It is believed that this constancy is achieved through multiple internal regulation and control processes. Given observations of a system, or even a detailed model of one, it is both valuable and extremely challenging to extract the control objectives of the homeostatic mechanisms. In this work, we develop a robust data-driven method to identify these objectives, namely to understand: "what does the system care about?". We propose an algorithm, Identifying Regulation with Adversarial Surrogates (IRAS), that receives an array of temporal measurements of the system and outputs a candidate for the control objective, expressed as a combination of observed variables. IRAS is an iterative algorithm consisting of two competing players. The first player, realized by an artificial deep neural network, aims to minimize a measure of invariance we refer to as the coefficient of regulation. The second player aims to render the task of the first player more difficult by forcing it to extract information about the temporal structure of the data, which is absent from similar "surrogate" data. We test the algorithm on four synthetic and one natural data set, demonstrating excellent empirical results. Interestingly, our approach can also be used to extract conserved quantities, e.g., energy and momentum, in purely physical systems, as we demonstrate empirically.

Twenty thousand leagues under plant biominerals: a deep learning implementation for automatic phytolith classification

Phytoliths constitute microscopic SiO2-rich biominerals formed in the cellular system of many living plants and are often preserved in soils, sediments and artefacts. Their analysis contributes significantly to the identification and study of botanical remains in (paleo)ecological and archaeological contexts. Traditional identification and classification of phytoliths rely on human experience, and as such, an emerging challenge is to automatically classify them to enhance data homogeneity among researchers worldwide and facilitate reliable comparisons. In the present study, a deep artificial neural network (NN) is implemented under the objective to detect and classify phytoliths, extracted from modern wheat (Triticum spp.). The proposed methodology is able to recognise four phytolith morphotypes: (a) Stoma, (b) Rondel, (c) Papillate, and (d) Elongate dendritic. For the learning process, a dataset of phytolith photomicrographs was created and allocated to training, validation and testing data groups. Due to the limited size and low diversity of the dataset, an end-to-end encoder-decoder NN architecture is proposed, based on a pre-trained MobileNetV2, utilised for the encoder part and U-net, used for the segmentation stage. After the parameterisation, training and fine-tuning of the proposed architecture, it is capable to classify and localise the four classes of phytoliths in unknown images with high unbiased accuracy, exceeding 90%. The proposed methodology and corresponding dataset are quite promising for building up the capacity of phytolith classification within unfamiliar (geo)archaeological datasets, demonstrating remarkable potential towards automatic phytolith analysis.

The Use of Artificial Neural Networks in the Public Sector

Artificial intelligence (AI) is an extensive scientific field, part of which is the concept of deep learning, belonging to broader family of machine learning (ML) methods, based on artificial neural networks (ANNs). ANNs are active since the 1940s and are applied in many fields. There have been actions around the world for the digital transformation of the public sector and the use of new innovative technologies, but the trajectory and degree of adoption of artificial intelligence technologies in the public sector have been unsatisfactory. Similar issues must be handled, and these problems must be classified. In the present work, preparatory searches were made on Scopus and IEEE bibliographic databases in order to obtain information for the progress of the adoption of ANNs in the public sector starting from the year 2019. Then, a systematic review of published scientific articles was conducted using keywords. Among the 2412 results returned by the search and the application of the selection/rejection criteria, 10 articles were chosen for analysis. The conclusion that emerged after reading the articles was that while the scientific community has a lot of suggestions and ideas for the implementation of ANNs and their financial effects, in practice, there is no appropriate use of them in the public sector. Occasionally, there are cases of implementation funded by state or non-state bodies without a systematic application and utilization of these technologies. The ways and methods of practical application are not further specified, so there are no indications for the systematic application of specialized deep learning techniques and ANNs. The legal framework for the development of artificial intelligence applications, at least in the European Union (EU), is under design, like the necessary ISO standards from an international perspective, and the economic impact of the most recent AI-based technologies has not been fully assessed.

Lower bounds for artificial neural network approximations: A proof that shallow neural networks fail to overcome the curse of dimensionality

Artificial neural networks (ANNs) have become a very powerful tool in the approximation of high-dimensional functions. Especially, deep ANNs, consisting of a large number of hidden layers, have been very successfully used in a series of practical relevant computational problems involving high-dimensional input data ranging from classification tasks in supervised learning to optimal decision problems in reinforcement learning. There are also a number of mathematical results in the scientific literature which study the approximation capacities of ANNs in the context of high-dimensional target functions. In particular, there are a series of mathematical results in the scientific literature which show that sufficiently deep ANNs have the capacity to overcome the curse of dimensionality in the approximation of certain target function classes in the sense that the number of parameters of the approximating ANNs grows at most polynomially in the dimension d∈N of the target functions under considerations. In the proofs of several of such high-dimensional approximation results it is crucial that the involved ANNs are sufficiently deep and consist a sufficiently large number of hidden layers which grows in the dimension of the considered target functions. It is the topic of this work to look a bit more detailed to the deepness of the involved ANNs in the approximation of high-dimensional target functions. In particular, the main result of this work proves that there exists a concretely specified sequence of functions which can be approximated without the curse of dimensionality by sufficiently deep ANNs but which cannot be approximated without the curse of dimensionality if the involved ANNs are shallow or not deep enough.

Deep Learning in High Dimension: Neural Network Expression Rates for Analytic Functions in \(\pmb{L^2(\mathbb{R}^d,\gamma_d)}\)

For artificial deep neural networks, we prove expression rates for analytic functions in the norm of where . Here denotes the Gaussian product probability measure on . We consider in particular and activations for integer . For , we show exponential convergence rates in . In case , under suitable smoothness and sparsity assumptions on , with denoting an infinite (Gaussian) product measure on , we prove dimension-independent expression rate bounds in the norm of . The rates only depend on quantified holomorphy of (an analytic continuation of) the map to a product of strips in (in for , respectively). As an application, we prove expression rate bounds of deep -NNs for response surfaces of elliptic PDEs with log-Gaussian random field inputs.

Machine learning as a new tool in neurological disease prevention, diagnosis, and treatment

More than 600 different neurological diseases affect the human population. Some of these are genetic and can emerge even before birth, and some are caused by defects, infections, trauma, degeneration, inflammation, and cancer. However, they all share disabilities caused by damage to the nervous system. In the last decades, the burden of almost all neurological disorders has increased in terms of absolute incidence, prevalence, and mortality, largely due to the population’s growth and aging. This represents a dangerous trend and should become our priority for the future. But what new goals are we going to set and reach now, and how will we exploit thought-provoking technological skills for making these goals feasible? Machine learning can be at the root of the problem. Indeed, most recently, there has been a push towards medical data analysis by machine learning, and a great improvement in the training capabilities particularly of artificial deep neural networks (DNNs) inspired by the biological neural networks characterizing the human brain. This has generated competitive results for applications such as biomolecular target and protein structure prediction, structure-based rational drug design, and repurposing, all exerting a major impact on neuroscience and human well-being. By approaching early risks for diseases, non-invasive diagnosis, personalized treatment assessment, drug discovery, and automated science, the machine learning arena has thus the potential of becoming the new frontier for empowering neuroscience research and clinical practice in the years ahead.

Robust estimation of shear wave velocity in a carbonate oil reservoir from conventional well logging data using machine learning algorithms

In recent decades, various applications of shear wave velocity have been reported in oil and gas projects. These applications include the determination of lithology, estimation of geomechanical parameters, reservoir fluid detection, etc. The shear wave velocity is usually measured or recorded by DSI (Dipole Shear Sonic imager) instruments. This log is not usually available in all wells of a hydrocarbon field because of the high cost of running this log compared to other conventional logs. However, it is tried to estimate the shear wave velocity from other related logs.The literature contains many empirical rock physics models used to estimate shear wave velocity. Most of them were developed for sandstone lithologies and didn't applicable in carbonate reservoir because of existing some vertical and horizontally complex fracture networks in the carbonate reservoirs that cause a quite different behavior for the shear wave velocity.The main objective of this research is to concentrate on the effect of lithology in the estimation of the target log for a carbonate reservoir. At the earliest step, preprocessing is performed and 455 data points of several conventional well logs, i.g., acoustic, density, neutron porosity, resistivity, gamma-ray, calcite volume, dolomite volume, and water saturation logs have been applied to generate the synthetic shear wave velocity log in a carbonate oil reservoir.First, original data are divided into training, validation, and testing subsets. Different data-driven predictive models were built: Multiple Linear Regression (MLR), Ensemble Learning Method (ELM), and three architectures of Artificial Neural Networks (ANN). The performance of all techniques was compared with each other. Finally, the feed-forward neural network shows the highest accuracy with R-values of 0.99 and 0.96 for the training and testing datasets. Moreover, the network design and tuning parameters of the model are adjusted using the grid search optimization technique to obtain the optimal design of the artificial neural network for estimating the shear wave velocity log. Hence, a deep artificial neural network, i.e., [8–11–15-1] network with bayesian regularization training algorithm and hyperbolic tangent sigmoid transfer function, is proposed to predict the target response in other wells.

Using Deep Learning with Bayesian–Gaussian Inspired Convolutional Neural Architectural Search for Cancer Recognition and Classification from Histopathological Image Frames

We propose a neural architectural search model which examines histopathological images to detect the presence of cancer in both lung and colon tissues. In recent times, deep artificial neural networks have made tremendous impacts in healthcare. However, obtaining an optimal artificial neural network model that could yield excellent performance during training, evaluation, and inferencing has been a bottleneck for researchers. Our method uses a Bayesian convolutional neural architectural search algorithm in collaboration with Gaussian processes to provide an efficient neural network architecture for efficient colon and lung cancer classification and recognition. The proposed model learns by using the Gaussian process to estimate the required optimal architectural values by choosing a set of model parameters through the exploitation of the expected improvement (EI) values, thereby minimizing the number of sampled trials and suggesting the best model architecture. Several experiments were conducted, and a landmark performance was obtained in both validation and test data through the evaluation of the proposed model on a dataset consisting of 25,000 images of five different classes with convergence and F1-score matrices.

High-resolution peak demand estimation using generalized additive models and deep neural networks

This paper covers predicting high-resolution electricity peak demand features given lower-resolution data. This is a relevant setup as it answers whether limited higher-resolution monitoring helps to estimate future high-resolution peak loads when the high-resolution data is no longer available. That question is particularly interesting for network operators considering replacing high-resolution monitoring by predictive models due to economic considerations. We propose models to predict half-hourly minima and maxima of high-resolution (every minute) electricity load data while model inputs are of a lower resolution (30 min). We combine predictions of generalized additive models (GAM) and deep artificial neural networks (DNN), which are popular in load forecasting. We extensively analyze the prediction models, including the input parameters’ importance, focusing on load, weather, and seasonal effects. The proposed method won a data competition organized by Western Power Distribution, a British distribution network operator. In addition, we provide a rigorous evaluation study that goes beyond the competition frame to analyze the models’ robustness. The results show that the proposed methods are superior to the competition benchmark concerning the out-of-sample root mean squared error (RMSE). This holds regarding the competition month and the supplementary evaluation study, which covers an additional eleven months. Overall, our proposed model combination reduces the out-of-sample RMSE by 57.4% compared to the benchmark.

Developing GAN-boosted Artificial Neural Networks to model the rate of drilling bit penetration

The goal of achieving a single model for estimating the rate of drilling bit penetration (ROP) with high accuracy has been the subject of many efforts. Analytical methods and, later, data-based techniques were utilized for this purpose. However, despite their partial effectiveness, these methods were inadequate for establishing models with sufficient generality. Based on deep learning (DL) concepts, this study has developed an innovative approach that produces general and boosted models capable of more accurately estimating ROPs compared to other techniques. A vital component of this approach is using a deep Artificial Neural Network (ANN) structure known as the Generative Adversarial Network (GAN). The GAN structure is combined with regressor (predictive) ANNs, to boost their performance when estimating the target parameter. The predictive ANNs of this study include Multi-Layer Perceptron Neural Network (MLP-NN) and 1-Dimensional Convolution Neural Network (1D-CNN) structures. More specifically, the key idea of our approach is to utilize GAN’s capability to produce fake samples comparable to true samples in predictive ANNs. Therefore, the proposed approach introduces a two-step predictive model development procedure. As the first step, the GAN structure is trained with the target of the problem as the input feature. GAN’s generator part, which can produce fake ROP samples similar to the true ones after training, is frozen and then replaces the output layer’s neuron of the predictive ANNs. In the second step, final predictive ANNs carrying frozen trained-generator called GAN-Boosted Neural Networks (GB-NNs) are trained to make predictions. Because this approach reduces the computational load of the predictive model training process and increases its quality, the performance of predictive ANN models is improved. An additional innovation of this research is using the residual structure during 1D-CNN network training, which improved the performance of the 1D-CNN by combining the input data with those features extracted from the inputs. This study revealed that the GB-Res 1D-CNN model, a GAN-Boosted 1-Dimensional Convolutional Neural Network with a Residual structure, results in the most accurate prediction. The validity of the GB-Res 1D-CNN model is confirmed by its successful implementation in blind well. As the final step of this study, we conducted a sensitivity analysis to identify the effect of different parameters on the predicted ROP. As expected, the DS and DT parameters significantly affect the model-estimated ROP.

PREVENTION STRATEGY OF MAIN NON-COMMUNICABLE DISEASES USING ARTIFICIAL NEURAL NETWORKS

ABSTRACT: The aim of this work was to build four models for the prediction of Noncommunicable Diseases (NCD) to support early diagnosis by applying deep Artificial Neural Networks (ANN) and clustering; through comparative analysis, the best paradigm was chosen. Following the methodology for data science, the objectives, project requirements and the quality of the data were analyzed, cleaning and normalization was carried out, the identification of the optimal number of groups by means of the silhouette method. Each model was trained with 70% of records and validated with the remaining 30% using the confusion matrix and F1. The comparative analysis showed the best performance of the multilayer perceptron with deep learning (over a network with clustering radial basis functions), its accuracy, precision, sensitivity, specificity and F1 were 99%, 97%, 100%, 97% and 98% respectively for lung cancer, 99%, 98%, 100%, 100% and 98% for breast cancer, Alzheimer with 98.6%, 100%, 96%, 100% and 97%, and depression with 91% , 87%, 93%, 88% and 88%, these models constitute an automated solution to strengthen medical diagnosis, they are considered a strategic support in the prevention of NCD, will contribute to improving the prognosis and quality of life of vulnerable people. Keywords: Prediction, artificial neural networks, data science, physical disorders, mental disorders