Sigmoid Transfer Function Research Articles

Early Recognition of Rainfall to Plan and Manage the Cultivation: A Forecasting Model Using Artificial Neural Networks for Dompe Divisional Secretariat in Gampaha District, Sri Lanka

As the agricultural productivity is climate sensitive, forecasting climatic factors is important to maximize the harvest and to planning and management of the cultivation. Even though all the climatic factors have influence on cultivation, rainfall is one of the major influential factor to Sri Lankan agriculture. Rainfall forecasting is vital in agriculture and it is a challenging task due to the uncertainty of natural phenomena. Artificial neural network (ANN) approach is being applied for forecasting real time rainfall using climatic factors affecting to rainfall. The aim of this study was to identify a neural network model which is capable of forecasting rainfall of Dompe Divisional Secretariat in Gampaha District with a high accuracy. Feed forward neural network model (FFNN) was applied with Levenberg-Marquardt back propagation algorithm and the network parameters were adjusted to minimize the forecasting error. The suggested FFNN model consisted of 12 input variables with two hidden layers with 13 and 10 hidden neurons in first and second layers, respectively. Log sigmoid transfer function used in hidden layer 1 and 2 while pure linear transfer function was used in the output layer. The model forecasts rainfall with mean squared error of 0.0895 and normalized mean squared error of 0.1975. The coefficient of determination (R²) of the testing set was 0.8. These results demonstrated the suitability of other climatic factors: temperature, wind speed, air pressure, humidity, percentage of clouds and their lags in forecasting rainfall using ANN technique in forecasting rainfall with high accuracy.

Artificial neural network based determination of the performance and emissions of a Diesel engine using ethanol-diesel fuel blends

In this study, the performance and exhaust emission values in a four-stroke, four-cylinder turbocharged Diesel engine fueled with ethanol-diesel fuel blends (10% and 15% in volume) were investigated by using Artificial Neural Network (ANN) modeling. The actual data derived from engine test measurements was applied in model training, cross-validation, and testing. To train the network, fuel injection pressures, throttle positions, engine speed, and ethanol fuel blend ratios were used as input layer in the network. The outputs are the engine performance values (engine torque, power, brake mean effective pressure, and specific fuel consumption) and exhaust emissions (SO2, CO2, NOx, and smoke level (N%)) which were measured in the experiments.The back-propagation learning algorithm with three different variants, a single layer, and logistic sigmoid transfer function (log-sig) was used in the network. By using the weights of the network, formulations were given for each output. The network for test data yielded the R2 values of 0.999 and the mean % errors for test data are smaller than 3.5% for the performance and 8% for the emissions.

Artificial neural network modelling for biodecolorization of Basic Violet 03 from aqueous solution by biochar derived from agro-bio waste of groundnut hull: Kinetics and thermodynamics

In this study, Levenberg Marquardt back propagation algorithm was used to train the Artificial Neural Network (ANN) and to predict the adsorptive removal of cationic dye Basic Violet 03 (BV03) by biochar derived from biowaste of groundnut hull. The experimental conditions such as solution pH, biochar dose, initial dye concentration, contact time and temperature were used as input variables and BV03 percentage removal as target. The hidden and the output layer of the network was trained by tangent sigmoid and liner transfer functions. The feasibility of the adsorption process is evaluated by the kinetic studies and it exhibited that pseudo-second order kinetic models fit well with experimental data. The adsorbent stability and adsorption mechanism has been discoursed by the thermodynamic characteristics and sorption free energy. The predicted target values were compared with the experiment resulted in a better correlation coefficient of 0.9920. Thus, the results attained from this ANN model was found to be effective in predicting the percentage removal of BV03 dye at any given operating condition.

Optimised ANN and SVR models for online prediction of moisture content and temperature of lentil seeds in a microwave fluidised bed dryer

Online monitoring and control of the drying processes are necessary to maintain the final products’ quality attributes, especially when a microwave system is used to facilitate the drying process. Machine learning techniques could be a suitable and very accurate approach for modelling the drying process. Two machine learning techniques including Support Vector Regression (SVR) and Artificial Neural Network (ANN) were employed to predict lentil seeds’ temperature and moisture ratio during drying in a microwave fluidised bed dryer with inputs of microwave power (0–500 W), fluidising air temperature (50 °C and 60 °C) and drying time. Mean squared error (MSE) and the coefficient of determination (R2) were used to evaluate the performance of the models. One hidden layer and 10 hidden neurons, the logistic sigmoid transfer function for the hidden layer, and a linear function for the output layer were determined to be the best structure for the ANN. Weights and biases were found by training the network with Bayesian Regularisation (trainbr) and the optimum MSE and R2 of the test set were 3.8×10-5 and 0.999 respectively. Optimised SVR could also provide a good model to predict MR and temperature (overall MSE = 1.96 ×10-4 andR2 = 0.995), although ANN provided relatively more accuracy for the temperature data. Therefore, ANN could be utilised as an accurate tool for the prediction of MR and temperature of lentil seeds during microwave fluidised bed drying.

Modeling of lead removal by living Scenedesmus obliquus using backpropagation (BP) neural network algorithm

Abstract Lead pollution in aquatic environment possessed lethal threats to public health. Heavy metal removal using microalgae have gained increasing attention as a novel biosorbent with great economic potential. However, excessive time consumption has been a limiting factor in algal adsorption experiments, while artificial neural network (ANN) prediction models could significantly improve the efficiency for experimental research. In this study, backpropagation (BP) neural network model was established for Pb 2 + removal from water by Scenedesmus obliquus with the 4:9:1 (input layer–hidden layer–output layer) construction. The experimental data of 275 groups derived from orthogonal experiments were obtained for training, validation and test data as input layer. Tangent sigmoid transfer function (tansig) was used in the hidden layer, while linear transfer function was used on the output layer (purelin). The correlation coefficient R2 reached 0.997 for the entire data set after training. Testing model results showed that the correlation between the predicted data and the experimental data was 0.997, with an accuracy rate of 95.4%, coefficient of determination of 0.999, root mean square error of 0.226, relative error of 1.47% and P value of 0.004. The predicted data of BP neural network also successfully applied in the pseudo-second-order kinetic model and Langmuir thermal dynamic model with the relative error less than 5%. Results showed that under pH 6 and algal biomass 2 g/L, Scenedesmus Obliquus achieved highest Pb 2 + removal efficiency. The successful application of BP neural network in Pb 2 + removal by Scenedesmus Obliquus provided an alternative and a more efficient method for microalgal adsorption experimental design and data validation.

Multi-Scale Extension in an Entorhinal-Hippocampal Model for Cognitive Map Building.

Neuroscience research shows that, by relying on internal spatial representations provided by the hippocampus and entorhinal cortex, mammals are able to build topological maps of environments and navigate. Taking inspiration from mammals' spatial cognition mechanism, entorhinal-hippocampal cognitive systems have been proposed for robots to build cognitive maps. However, path integration and vision processing are time-consuming, and the existing model of grid cells is hard to achieve in terms of adaptive multi-scale extension for different environments, resulting in the lack of viability for real environments. In this work, an optimized dynamical model of grid cells is built for path integration in which recurrent weight connections between grid cells are parameterized in a more optimized way and the non-linearity of sigmoidal neural transfer function is utilized to enhance grid cell activity packets. Grid firing patterns with specific spatial scales can thus be accurately achieved for the multi-scale extension of grid cells. In addition, a hierarchical vision processing mechanism is proposed for speeding up loop closure detection. Experiment results on the robotic platform demonstrate that our proposed entorhinal-hippocampal model can successfully build cognitive maps, reflecting the robot's spatial experience and environmental topological structures.

Hybrid Binary Grey Wolf With Harris Hawks Optimizer for Feature Selection

Despite Grey Wolf Optimizer’s (GWO) superior performance in many areas, stagnation in local optima areas may still be a concern. Several significant GWO factors can be explored to enhance the performance of selection in classification, with two conflicting concepts to be considered in using or modeling a metaheuristic method, exploring a search field, and exploiting optimal solutions. Balancing exploration and exploitation in a good manner will improve the search algorithm’s performance. To achieve a good balance, this paper proposes a binary hybrid GWO and Harris Hawks Optimization (HHO) to form a memetic approach called HBGWOHHO. The sigmoid transfer function is used to transfer the continuous search space into a binary one to meet the feature selection nature requirement. A wrapper-based k-Nearest neighbor is used to evaluate the goodness of the selected features. To validate the performance of the proposed method, 18 standard UCI benchmark datasets were used. The performance of the proposed hybrid method was compared with Binary Grey Wolf Optimizer (BGWO), Binary Particle Swarm Optimization (BPSO), Binary Harris Hawks Optimizer (BHHO), Binary Genetic Algorithm (BGA) and Binary Hybrid BWOPSO. The findings revealed that the proposed method was effective in improving the performance of the BGWO algorithm. The proposed hybrid method outperforms the BGWO algorithm in terms of accuracy, selected feature size, and computational time. Similarly, compared with BPSO and BGA feature selection algorithms, the proposed HBGWOHHO surpassed them yield better accuracy, the smaller size of selected features in much lower computational time.

Response Surface Methodology and Artificial Neural Network Methods Comparative Assessment for Fuel Rich and Fuel Lean Catalytic Combustion

Modeling, predictive and generalization capabilities of response surface methodology (RSM) and artificial neural network (ANN) have been performed to assess the thermal structure of the experimentally studied catalytic combustion of stabilized confined turbulent gaseous diffusion flames. The Pt/γAl2O3 and Pd/γAl2O3 disc burners were located in the combustion domain and the experiments were accomplished under both fuel-rich and fuel-lean conditions at a modified equivalence (fuel/air) ratio (ø) of 0.75 and 0.25, respectively. The thermal structure of these catalytic flames developed over the Pt and Pd disc burners was scrutinized via measuring the mean temperature profiles in the radial direction at different discrete axial locations along with the flames. The RSM and ANN methods investigated the effect of the two operating parameters namely (r), the radial distance from the center line of the flame, and (x), axial distance along with the flame over the disc, on the measured temperature of the flames and predicted the corresponding temperatures beside predicting the maximum temperature and the corresponding input process variables. A three-layered Feed Forward Neural Network was developed in conjugation with the hyperbolic tangent sigmoid (tansig) transfer function and an optimized topology of 2:10:1 (input neurons:hidden neurons:output neurons). Also the ANN method has been exploited to illustrate the effects of coded R and X input variables on the response in the three and two dimensions and to locate the predicted maximum temperature. The results indicated the superiority of ANN in the prediction capability as the ranges of & F_Ratio are 0.9181 - 0.9809 & 634.5 - 3528.8 for RSM method compared to 0.9857 - 0.9951 & 7636.4 - 24,028.4 for ANN method beside lower values for error analysis terms.

Detecting Depression in Elderly People by Using Artificial Neural Network

Introduction: The possibility of depression is common in the elderly. Novel technologies allow us to monitor people related to depression. Hence, a model was provided to detect depression in elderly based on artificial neural network (ANN). Methods: The present study is an applied descriptive-survey research. Forty elderly people were randomly selected from the Elderly Care Center in Gonbad Kavous, Golestan, Iran in 2019. Data were obtained through interview. The data were randomly divided in to two groups of training and testing. In training phase by using first dataset (70%), three layers network is considered. Interrelation weights between variables, optimum transfer function and optimum number of hidden layer were obtained. The sum of squared errors, receiver operating characteristic curve criterion and accuracy were used to select the optimum ANN. The optimum model tested and validated (p < 0.001) with second dataset (30%). Results: The sigmoid transfer function in hidden and output layers with 5 nods (SSE = 131), one hidden layer with 15 neurons was considered as optimum model. Receiver operating characteristic curve criterion and accuracy were obtained equal to 0.913 and 94.79% respectively. The confusion matrix was showed high sensitivity (97.45%) and specificity (99.25%) in the diagnosis of depression. Age, gender, income, polarity outgoing messages to family, incoming calls, time active in the day, polarity incoming messages from family, time sleep in the day were obtained as a significant set for input layer of the optimum model. In addition, the optimum model has been quite successful in identifying normal and depressed elderly. Conclusion: This research applied an ANN model for detection of depression in the elderly. ANN can be used as a computational tools for early diagnosis of depression in the elderly People.

Dynamic Causal Modeling for fMRI With Wilson-Cowan-Based Neuronal Equations.

Dynamic causal modeling (DCM) is an analysis technique that has been successfully used to infer about directed connectivity between brain regions based on imaging data such as functional magnetic resonance imaging (fMRI). Most variants of DCM for fMRI rely on a simple bilinear differential equation for neural activation, making it difficult to interpret the results in terms of local neural dynamics. In this work, we introduce a modification to DCM for fMRI by replacing the bilinear equation with a non-linear Wilson-Cowan based equation and use Bayesian Model Comparison (BMC) to show that this modification improves the model evidences. Improved model evidence of the non-linear model is shown for our empirical data (imitation of facial expressions) and validated by synthetic data as well as an empirical test dataset (attention to visual motion) used in previous foundational papers. For our empirical data, we conduct the analysis for a group of 42 healthy participants who performed an imitation task, activating regions putatively containing the human mirror neuron system (MNS). In this regard, we build 540 models as one family for comparing the standard bilinear with the modified Wilson-Cowan models on the family-level. Using this modification, we can interpret the sigmoid transfer function as an averaged f-I curve of many neurons in a single region with a sigmoidal format. In this way, we can make a direct inference from the macroscopic model to detailed microscopic models. The new DCM variant shows superior model evidence on all tested data sets.

Dynamic Modeling Using Artificial Neural Network of Bacillus Velezensis Broth Cross-Flow Microfiltration Enhanced by Air-Sparging and Turbulence Promoter.

Cross-flow microfiltration is a broadly accepted technique for separation of microbial biomass after the cultivation process. However, membrane fouling emerges as the main problem affecting permeate flux decline and separation process efficiency. Hydrodynamic methods, such as turbulence promoters and air sparging, were tested to improve permeate flux during microfiltration. In this study, a non-recurrent feed-forward artificial neural network (ANN) with one hidden layer was examined as a tool for microfiltration modeling using Bacillus velezensis cultivation broth as the feed mixture, while the Kenics static mixer and two-phase flow, as well as their combination, were used to improve permeate flux in microfiltration experiments. The results of this study have confirmed successful application of the ANN model for prediction of permeate flux during microfiltration of Bacillus velezensis cultivation broth with a coefficient of determination of 99.23% and absolute relative error less than 20% for over 95% of the predicted data. The optimal ANN topology was 5-13-1, trained by the Levenberg–Marquardt training algorithm and with hyperbolic sigmoid transfer function between the input and the hidden layer.

Discrimination of cycling patterns using accelerometric data and deep learning techniques

The monitoring of physical activities and recognition of motion disorders belong to important diagnostical tools in neurology and rehabilitation. The goal of the present paper is in the contribution to this topic by (1) analysis of accelerometric signals recorded by wearable sensors located at specific body positions and by (2) implementation of deep learning methods to classify signal features. This paper uses the general methodology to analysis of accelerometric signals acquired during cycling at different routes followed by the global positioning system. The experimental dataset includes 850 observations that were recorded by a mobile device in the spine area (L3 verterbra) for cycling routes with the different slope. The proposed methodology includes the use of deep learning convolutional neural networks with five layers applied to signal values transformed into the frequency domain without specification of any signal features. The accuracy of discrimination between different motion patterns for the uphill and downhill cycling and recognition of 4 classes associated with different route slopes was 96.6% with the loss criterion of 0.275 for sigmoidal activation functions. These results were compared with those evaluated for selected sets of features estimated for each observation and classified by the support vector machine, Bayesian methods, and the two-layer neural network. The best cross-validation error of 0.361 was achieved for the two-layer neural network model with the sigmoidal and softmax transfer functions. Our methodology suggests that deep learning neural networks are efficient in the assessment of motion activities for automated data processing and have a wide range of applications, including rehabilitation, early diagnosis of neurological problems, and possible use in engineering as well.

Predict the Effects of Forming Tool Characteristics on Surface Roughness of Aluminum Foil Components Formed by SPIF Using ANN and SVR

In the present work, multiple forming tests were conducted under different forming conditions by Single Point Incremental Forming (SPIF). In which surface roughness, arithmetical mean roughness (Ra) and the ten-point mean roughness (Rz) of AlMn1Mg1 sheet were experimentally measured. Also, an Artificial Neural Network (ANN) was used to predict the (Ra) and (Rz) by adopting the data collected from 108 components that were formed by SPIF. Forming tool characteristics played a key role in all the predictions and their effect on the final product surface roughness. In the aim to explore the proper materials and geometry of forming tools, different ANN structures, different training, and transfer functions have been applied to predict (Ra) and (Rz) as an output argument. Furthermore, Support Vector Regression (SVR) with different kernel types have been used for prediction, together with Gradient Boosting regression to sort the effective parameters on the surface roughness. The input arguments were tool materials, tool shape, tool end/corner radius, and tool surface roughness (Ra and Rz). The actual data subjected to a fit regression model to generate prediction equations of Ra and Rz. The results showed that ANN with one output gives the best R-Square (R2). Levenberg-Marquardt backpropagation (Trainlm) training function recorded the highest value of R2, 0.9628 for prediction Ra using Softmax transfer function whereas 0.9972 for Rz by Log- Sigmoid transfer function. Furthermore, tool materials, together with tool surface (Ra), are playing a significant importance role, affecting the sheet surface roughness (Ra). Whereas tool roughness Rz was the critical parameter effected on the Rz of the product. Also, there was a significant positive effect of tool geometry on the sheet surface roughness.

A new approach for the vanishing gradient problem on sigmoid activation

The vanishing gradient problem (VGP) is an important issue at training time on multilayer neural networks using the backpropagation algorithm. This problem is worse when sigmoid transfer functions are used, in a network with many hidden layers. However, the sigmoid function is very important in several architectures such as recurrent neural networks and autoencoders, where the VGP might also appear. In this article, we propose a modification of the backpropagation algorithm for the sigmoid neurons training. It consists of adding a small constant to the calculation of the sigmoid’s derivative so that the proposed training direction differs slightly from the gradient while keeping the original sigmoid function in the network. This approach suggests that the derivative’s modification produces the same accuracy in fewer training steps on most datasets. Moreover, due to VGP, the original derivative does not converge using sigmoid functions on more than five hidden layers. However, the modification allows backpropagation to train two extra hidden layers in feedforward neural networks.

Binary Whale Optimization Algorithm for Dimensionality Reduction

Feature selection (FS) was regarded as a global combinatorial optimization problem. FS is used to simplify and enhance the quality of high-dimensional datasets by selecting prominent features and removing irrelevant and redundant data to provide good classification results. FS aims to reduce the dimensionality and improve the classification accuracy that is generally utilized with great importance in different fields such as pattern classification, data analysis, and data mining applications. The main problem is to find the best subset that contains the representative information of all the data. In order to overcome this problem, two binary variants of the whale optimization algorithm (WOA) are proposed, called bWOA-S and bWOA-V. They are used to decrease the complexity and increase the performance of a system by selecting significant features for classification purposes. The first bWOA-S version uses the Sigmoid transfer function to convert WOA values to binary ones, whereas the second bWOA-V version uses a hyperbolic tangent transfer function. Furthermore, the two binary variants introduced here were compared with three famous and well-known optimization algorithms in this domain, such as Particle Swarm Optimizer (PSO), three variants of binary ant lion (bALO1, bALO2, and bALO3), binary Dragonfly Algorithm (bDA) as well as the original WOA, over 24 benchmark datasets from the UCI repository. Eventually, a non-parametric test called Wilcoxon’s rank-sum was carried out at 5% significance to prove the powerfulness and effectiveness of the two proposed algorithms when compared with other algorithms statistically. The qualitative and quantitative results showed that the two introduced variants in the FS domain are able to minimize the selected feature number as well as maximize the accuracy of the classification within an appropriate time.

Yapay Sinir Ağı Yaklaşımı ile Atık Portakal Kabuğundan Elde Edilen Grafen Benzeri Gözenekli Karbon Üzerinde Arsenik (V) Biyosorpsiyonunun Modellenmesi

The main threats from heavy metals specifically arsenic-contaminated drinking water have been emerging as an environmental and social crucial issue. Herein, the arsenic (V) (As(V)) biosorption performance of waste orange peel (OP) driven-graphene-like porous carbon (GPC) was investigated experimentally and an artificial neural network (ANN) approach was used to model the biosorption process. The initial pH (2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, and 10.0), initial As(V) concentration (25.0, 50.0, 100.0, 250.0, 500.0 and 750.0 mg.L-1), biosorbent dosage (1.0, 2.0, 3.0, 4.0 and 5.0 g.L-1), and contact time (0– 120.0 min) were investigated to optimize the biosorption process. The as-synthesized GPC biosorbents with a high specific surface area (985 m2.g-1) and pore volume (1.04 cm3.g-1) offered superior removal efficiency as 88.2% (equilibrium uptake capacity of 46.5 mg.g-1) at initial pH 6.0, initial As(V) concentration 100 mg.L-1, and biosorbent dosage 2.0 g.L-1. A three-layer ANN model was developed to forecast the Ar(V) biosorption performance of GPCs. Several experimental data points were considered as test data to validate the ANN model. The ANN model was performed with the Levengberg-Marquardt algorithm (LMA), linear transfer function (purelin) at the output layer, and a tangent sigmoid transfer function (tansig) in the hidden layer with 12 neurons. The values of coefficient of determination and mean squared error were calculated to be 0.9858 and 0.0014, respectively. The results revealed that the experimental data were in accordance with ANN-driven data as well as reveling the high accuracy of the ANN approach in estimating the target variable. The developed ANN model is useful for the optimization of process conditions for pilot-scale utilization of As(V) biosorption process by GPC.

Application of artificial neural network in predicting EI.

IntroductionEmotional intelligence (EI) constitutes a whole set of non-cognitive capabilities, competencies, and skills that affect one's ability to deal successfully with environmental demands and pressures. Different factors such as gender, age, education, place of residence, etc. can influence this variable. Nevertheless, the influence of a multitude of factors involved in behavioral phenomena cannot often be controlled.PurposeTherefore, some difficulty may often raise in finding associations between these variables using regression models as regression models are built on restrictive assumptions.MethodsIn these cases, models such as artificial neural networks are excellent alternatives to regression models. In this study, the neural network model was used in SPSS software to predict the pattern held among the variables of age, gender, occupation, marital status, and education for predicting the EI of 901 individuals aged from 17 to 73 years.ResultsThe appropriate neural network model for EI prediction is a hyperbolic tangent transfer function with two neurons in the hidden layer and a sigmoid transfer function in the output layer. This network was able to predict EI in most of its dimensions with significant correlations and could demonstrate the neural network's advantage over regression models in predicting EI using sociological variables.ConclusionThis model is able to estimate the EI level in different occupational, educational, gender, and age groups, and provide the ground for planning to address potential deficiencies in each group.

Thermoeconomic modeling and artificial neural network optimization of Afyon geothermal power plant

The Afyon Geothermal Power Plant is modeled using the Multi-Layer Feed-Forward Artificial Neural Network. The 100 × 8 data set obtained from the real Binary Geothermal Power Plant is divided into two parts: 80 × 8 training data and 20 × 8 test data. Geothermal Power Plant system modeling has been performed numerically on Matlab with three inputs and five outputs. There are ten neurons in the hidden layer in the Artificial Neural Network-based system, and the logarithmic sigmoid transfer function is used as the transfer function in each neuron. The neurons in the output layer have the purelin transfer function. As a result of the training process, the 3.06 × 10E-2 mean square error value was obtained from the ANN-based Binary Geothermal Power Plant system. The main point of the study is the optimization of the binary geothermal power plant. The genetic algorithm method with Artificial Neural Network-based is used for this purpose. The results obtained from the outputs of the Artificial Neural Network-based Binary Geothermal Power Plant system are presented. The plant’s geothermal water temperature and mass flow rates are 110 °C and 150 kg/s. Energy and exergy efficiencies of the plant are calculated as 10.4% and 29.7%. The optimized simple payback period and exergy cost of the electricity generated in the plant is calculated as 2.87 years and 0.0176 $/kWh, respectively.

Prediction of frictional pressure losses in annulus with artificial neural network modeling

Pressure loss of single phase in annular flow for non-Newtonian drilling mud have been predicted utilizing an artificial neural network (ANN). Based on statistical study, an optimization procedure was performed to choose the optimum of ANN technique. With Levenberg-Marquardt back-propagation training algorithm, six variables as input and once as outcome were modeled and trained using a three-layer feed forward neural network. Three neurons, the log sigmoid transfer function in the hidden layer and the linear transfer function inside the result layer were introduced in this article. Simulated results of ANN technique are been demonstrated that there is an outstanding agreement among predicting values and measuring pressure drops in annulus for test data and training sets. Additionally, typical model used (power law model) that were applied to calculate the performance of ANN model and comparison the three wells results. Based on the analysis of the results obtained, it is shown that the ANN method outperforms other models significantly and delivers excellent predictions, the correlation coefficient (R2) of 0.9999 average absolute relative error (AARE) equal to 0.00001.

Machine learning approach for prediction of paracetamol adsorption efficiency on chemically modified orange peel.

High consumption of paracetamol (PCM) has led to the discharge of a large quantity of its metabolite into the environment and there is an urgent need to remove this harmful contaminant in a sustainable manner. In this work, Artificial Neural Network (ANN) was used as a Machine Learning tool for prediction of PCM adsorption efficiency on chemically modified orange peel (CMOP). Orange peel was chemically modified with orthophosphoric acid and then characterized using Scanning Electron Microscopy (SEM) and Fourier Transform Infrared Spectroscopy (FTIR). Thereafter, batch adsorption of PCM on CMOP were conducted at different operating conditions namely: contact time (0-330min), temperature (30-50°C) and initial drug concentration (10mg/L-50mg/L) to obtain the residual concentration of PCM in solution. Experimental data was used to compute the adsorption efficiency of PCM on CMOP. To predict the adsorption efficiency, different ANN architectures were examined. A neural network structure with Levenberg Marquardt (LM) training algorithm, 17 hidden neurons, and tangent sigmoid transfer function at both the input and output layers gave the best level of prediction. Comparing with experimental data, the optimal model yielded Mean Square Error (MSE), Root Mean Square Error (RMSE), and Correlation coefficient (R2) of 5.8985×10-04, 0.0243 and 0.9958 respectively. The results obtained showed that ANN is efficient in predicting the adsorption efficiency of PCM on CMOP.