Comparative study of ANN (artificial neural network) versus RSM (response surface methodology) for predicting the recovery of phenolic compounds from spent coffee grounds by conventional and microwave assisted extraction

Sravanthi Budaraju, Kumar Mallikarjunan

Research output: Contribution to conferencePaper

Abstract

The current study investigated the efficacy of RSM (response surface methodology) and ANN (artificial neural network) models to validate the data and to predict the maximum total phenolic content (TPC) extracted from spent coffee grounds by conventional and microwave assisted extraction (MAE) systems. Experimental conditions such as temperature (20oC, 40oC, 60oC), time (30 min, 60 min, 90 min), and sample mass (0.5 g, 1.0 g, 1.5 g) for solvent extraction and power (400 W, 800 W, 1200 W), extraction time (40 s, 80 s, 120 s) and sample mass (0.5 g, 1.0 g, 1.5 g) for MAE were considered. A central composite face-centered design has been employed to monitor the combined effect of extraction characteristics, and the results were analyzed using Design Expert Software for RSM and MATLAB R2017b for ANN. The data obtained from the experimental design was fitted to second-order polynomial response surface model which was applied to fit the experimental results obtained by face-centered design. A feed-forward MLP (Multilayer Perceptron) ANN with three or more layers of hidden neurons using backpropagation was used for the validation and testing of ANN-model. The results showed that 60oC temperature, 90 min time, 0.5 g of sample mass combination gave maximum results with 44.4 mg g-1 of a dry extract of phenolic for solvent extraction whereas, 966.3 W power, 49.0 s time and 0.5g of sample mass extracted 57.5 mg g-1 of a dry extract of phenolic compounds from MAE. MAE yielded higher TPC values than conventional solvent technique with lower time and higher yield. ANN trained network gave the maximum R2 value of 0.748 when compared to 0.999 for RSM and average absolute deviation (AAD) values of 2.368% versus 8.105% for RSM. ANN also showed a good agreement between the predicted and the actual TPC values for the selected extraction conditions. The current data shows that the RSM could be a useful mathematical tool to optimize the extraction process while the ANN is a better method for predicting TPC extracted from spent coffee grounds. These research findings could provide an effective guideline, and the results would be a good database to the food industry applications.

Original languageEnglish (US)
DOIs
StatePublished - Jan 1 2018
EventASABE 2018 Annual International Meeting - Detroit, United States
Duration: Jul 29 2018Aug 1 2018

Other

OtherASABE 2018 Annual International Meeting
CountryUnited States
CityDetroit
Period7/29/188/1/18

Fingerprint

Coffee
response surface methodology
neural networks
phenolic compounds
Microwaves
Neural networks
Recovery
Solvent extraction
Multilayer neural networks
sampling
Backpropagation
Design of experiments
MATLAB
Neurons
extracts
Polynomials
food industry
temperature
Temperature
neurons

Keywords

  • Artificial neural network
  • Microwave assisted extraction
  • Response surface methodology
  • Spent coffee grounds
  • Total phenolic content.

Cite this

Comparative study of ANN (artificial neural network) versus RSM (response surface methodology) for predicting the recovery of phenolic compounds from spent coffee grounds by conventional and microwave assisted extraction. / Budaraju, Sravanthi; Mallikarjunan, Kumar.

2018. Paper presented at ASABE 2018 Annual International Meeting, Detroit, United States.

Research output: Contribution to conferencePaper

@conference{be1bf2fb039e46f0be7d8e2a979132da,
title = "Comparative study of ANN (artificial neural network) versus RSM (response surface methodology) for predicting the recovery of phenolic compounds from spent coffee grounds by conventional and microwave assisted extraction",
abstract = "The current study investigated the efficacy of RSM (response surface methodology) and ANN (artificial neural network) models to validate the data and to predict the maximum total phenolic content (TPC) extracted from spent coffee grounds by conventional and microwave assisted extraction (MAE) systems. Experimental conditions such as temperature (20oC, 40oC, 60oC), time (30 min, 60 min, 90 min), and sample mass (0.5 g, 1.0 g, 1.5 g) for solvent extraction and power (400 W, 800 W, 1200 W), extraction time (40 s, 80 s, 120 s) and sample mass (0.5 g, 1.0 g, 1.5 g) for MAE were considered. A central composite face-centered design has been employed to monitor the combined effect of extraction characteristics, and the results were analyzed using Design Expert Software for RSM and MATLAB R2017b for ANN. The data obtained from the experimental design was fitted to second-order polynomial response surface model which was applied to fit the experimental results obtained by face-centered design. A feed-forward MLP (Multilayer Perceptron) ANN with three or more layers of hidden neurons using backpropagation was used for the validation and testing of ANN-model. The results showed that 60oC temperature, 90 min time, 0.5 g of sample mass combination gave maximum results with 44.4 mg g-1 of a dry extract of phenolic for solvent extraction whereas, 966.3 W power, 49.0 s time and 0.5g of sample mass extracted 57.5 mg g-1 of a dry extract of phenolic compounds from MAE. MAE yielded higher TPC values than conventional solvent technique with lower time and higher yield. ANN trained network gave the maximum R2 value of 0.748 when compared to 0.999 for RSM and average absolute deviation (AAD) values of 2.368{\%} versus 8.105{\%} for RSM. ANN also showed a good agreement between the predicted and the actual TPC values for the selected extraction conditions. The current data shows that the RSM could be a useful mathematical tool to optimize the extraction process while the ANN is a better method for predicting TPC extracted from spent coffee grounds. These research findings could provide an effective guideline, and the results would be a good database to the food industry applications.",
keywords = "Artificial neural network, Microwave assisted extraction, Response surface methodology, Spent coffee grounds, Total phenolic content.",
author = "Sravanthi Budaraju and Kumar Mallikarjunan",
year = "2018",
month = "1",
day = "1",
doi = "10.13031/aim.201800169",
language = "English (US)",
note = "ASABE 2018 Annual International Meeting ; Conference date: 29-07-2018 Through 01-08-2018",

}

TY - CONF

T1 - Comparative study of ANN (artificial neural network) versus RSM (response surface methodology) for predicting the recovery of phenolic compounds from spent coffee grounds by conventional and microwave assisted extraction

AU - Budaraju, Sravanthi

AU - Mallikarjunan, Kumar

PY - 2018/1/1

Y1 - 2018/1/1

N2 - The current study investigated the efficacy of RSM (response surface methodology) and ANN (artificial neural network) models to validate the data and to predict the maximum total phenolic content (TPC) extracted from spent coffee grounds by conventional and microwave assisted extraction (MAE) systems. Experimental conditions such as temperature (20oC, 40oC, 60oC), time (30 min, 60 min, 90 min), and sample mass (0.5 g, 1.0 g, 1.5 g) for solvent extraction and power (400 W, 800 W, 1200 W), extraction time (40 s, 80 s, 120 s) and sample mass (0.5 g, 1.0 g, 1.5 g) for MAE were considered. A central composite face-centered design has been employed to monitor the combined effect of extraction characteristics, and the results were analyzed using Design Expert Software for RSM and MATLAB R2017b for ANN. The data obtained from the experimental design was fitted to second-order polynomial response surface model which was applied to fit the experimental results obtained by face-centered design. A feed-forward MLP (Multilayer Perceptron) ANN with three or more layers of hidden neurons using backpropagation was used for the validation and testing of ANN-model. The results showed that 60oC temperature, 90 min time, 0.5 g of sample mass combination gave maximum results with 44.4 mg g-1 of a dry extract of phenolic for solvent extraction whereas, 966.3 W power, 49.0 s time and 0.5g of sample mass extracted 57.5 mg g-1 of a dry extract of phenolic compounds from MAE. MAE yielded higher TPC values than conventional solvent technique with lower time and higher yield. ANN trained network gave the maximum R2 value of 0.748 when compared to 0.999 for RSM and average absolute deviation (AAD) values of 2.368% versus 8.105% for RSM. ANN also showed a good agreement between the predicted and the actual TPC values for the selected extraction conditions. The current data shows that the RSM could be a useful mathematical tool to optimize the extraction process while the ANN is a better method for predicting TPC extracted from spent coffee grounds. These research findings could provide an effective guideline, and the results would be a good database to the food industry applications.

AB - The current study investigated the efficacy of RSM (response surface methodology) and ANN (artificial neural network) models to validate the data and to predict the maximum total phenolic content (TPC) extracted from spent coffee grounds by conventional and microwave assisted extraction (MAE) systems. Experimental conditions such as temperature (20oC, 40oC, 60oC), time (30 min, 60 min, 90 min), and sample mass (0.5 g, 1.0 g, 1.5 g) for solvent extraction and power (400 W, 800 W, 1200 W), extraction time (40 s, 80 s, 120 s) and sample mass (0.5 g, 1.0 g, 1.5 g) for MAE were considered. A central composite face-centered design has been employed to monitor the combined effect of extraction characteristics, and the results were analyzed using Design Expert Software for RSM and MATLAB R2017b for ANN. The data obtained from the experimental design was fitted to second-order polynomial response surface model which was applied to fit the experimental results obtained by face-centered design. A feed-forward MLP (Multilayer Perceptron) ANN with three or more layers of hidden neurons using backpropagation was used for the validation and testing of ANN-model. The results showed that 60oC temperature, 90 min time, 0.5 g of sample mass combination gave maximum results with 44.4 mg g-1 of a dry extract of phenolic for solvent extraction whereas, 966.3 W power, 49.0 s time and 0.5g of sample mass extracted 57.5 mg g-1 of a dry extract of phenolic compounds from MAE. MAE yielded higher TPC values than conventional solvent technique with lower time and higher yield. ANN trained network gave the maximum R2 value of 0.748 when compared to 0.999 for RSM and average absolute deviation (AAD) values of 2.368% versus 8.105% for RSM. ANN also showed a good agreement between the predicted and the actual TPC values for the selected extraction conditions. The current data shows that the RSM could be a useful mathematical tool to optimize the extraction process while the ANN is a better method for predicting TPC extracted from spent coffee grounds. These research findings could provide an effective guideline, and the results would be a good database to the food industry applications.

KW - Artificial neural network

KW - Microwave assisted extraction

KW - Response surface methodology

KW - Spent coffee grounds

KW - Total phenolic content.

UR - http://www.scopus.com/inward/record.url?scp=85054196655&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85054196655&partnerID=8YFLogxK

U2 - 10.13031/aim.201800169

DO - 10.13031/aim.201800169

M3 - Paper

ER -