Cowpea is widely grown and consumed in Sub-Saharan Africa because of its low cost and high mineral, protein, and other nutritional content. Nonetheless, cooking it takes considerable time, and there have been attempts on techniques for speeding up the cooking process without compromising its nutritious value. Infrared heating has recently been proposed as a viable way of preparing instantized cowpea grains that take a short amount of time to cook while maintaining desired sensory characteristics. Despite this, only a few studies have shown the impact of moisture, temperature, and cooking time on cooking characteristics such as bulk density, water absorption (WABS), and the pectin solubility of infrared heated cowpea precooked using this technology. Artificial neural network was used as a machine learning tool to study the effect of a prediction model on the infrared heating performance and cooking characteristics of precooked cowpea seeds. With R values of 0.987, 0.991, and 0.938 for the bulk density, WABS, and pectin solubility, respectively, the prediction model created in this study utilizing an artificial neural network (a type of machine learning) outperformed the traditional linear, 2-Factor Interaction (2FI), and quadratic models.
Research Article
Prediction of cooking characteristics of moisture conditioned and infrared heated cowpea using Artificial neural network
https://doi.org/10.21203/rs.3.rs-1411357/v1
This work is licensed under a CC BY 4.0 License
You are reading this latest preprint version
Cowpea
Infrared heating
Bulk density
Water absorption
Pectin solubility
Artificial neural network
Cowpea (Vigna unguiculata) is among the leguminous crop mostly cultivated in Sub-Saharan Africa [1]. It is an affordable and sustainable plant source of protein, phytonutrients, and minerals, useful in combating protein-energy malnutrition and food insecurity [2]. Cowpea remains a vital raw material used in making traditional dishes in various forms such as couscous (shô basi), fritters (akara), steamed pudding (moimoi)[3, 4]. Cowpea is mainly consumed by cooking the whole grains in water for about 2 h, however this long processing time and as a result high energy consumption during food preparation remains the major limitation to its consumption in urban areas [3]. Several methods including soaking, precooking, and usage of alkaline salts treatments have been useful to decrease cooking time and enhancing the nutritional quality of pulses [5–7]. Recently, infrared heating, a novel thermal food processing technique has also been reported to significantly improved cooking characteristics of pulses such as cowpea [8].
Infrared radiation heating is a hydrothermal process that uses electromagnetic radiation in the infra-red region within the wavelength of about 3 to 1000 µm to generate heat [9]. Infrared technology has been useful in many foods manufacturing processes, including drying, boiling, heating, roasting of food, cooking food and sterilization grains [10]. It is an environmentally friendly thermal treatment and considered as an advanced thermal process that can be exploited in food processing. Different studies on infrared heating of cowpea grains have shown that the combination of varying moisture level, infrared radiation temperature and time could reduce its cooking time by more than 50%, improve sensory properties including appearance and color and protein quality of the infrared treated cowpea grains [3, 8, 11, 12]. These processing conditions causes an increase in solubility of pectin responsible for rapid water uptakes, separation of cells along the cell wall and other physicochemical and structural changes of starch and protein as a result facilitated softening of cowpeas and reduction in the cooking time [8, 11]. Developing an optimization model has been useful in predicting and optimizing the parameters in grain processing to improve the pulses quality for instance cowpea grains [8, 13].
Response surface methodology (RSM) is one of the utilized conventional optimization model techniques in pulses processing that has been useful in predicting and optimizing processing parameters such as germination [14], drying [15], extrusion [16] and infrared heating [8, 13] of pulses to improve quality and developing new products. However, the performance of models developed using RSM is limited. Recently, with the advent of the 4th industrial revolution, artificial neural network has been considered as a more efficient tool for model predictions of biodiesel production [17] and even in food processing [18]. An artificial neural network (ANN) is a mathematical algorithm which has the capability of relating the input and output parameters, learning from examples through iteration, without requiring a prior knowledge of the relationships of the process parameters [18]. Artificial neural network models have been used for process control in thermal food processing including modelling tool in several food processing applications like drying [19, 20], heat transfer and thermal process predictions [18, 21]. Furthermore, artificial intelligence is one of the approaches which has proven to be efficient with improving the processing quality of grains like wheat in the development of food products [22, 23]. Table 1 shows some of the Application of ANN for food processes. Nonetheless, to the best of our knowledge, there is still a dearth of information available on predicting the properties of pulses prepared by infrared heating using artificial intelligence, which is critical for improving the cooking characteristics of pulses such as cowpea, as an important source of protein. Therefore, this study aimed at developing a predictive model for evaluating the cooking characteristics of cowpea (specifically bulk density, water absorption capacity and pectin solubility) under varied processing parameters of moisture content, temperature, and time within a closed system of an infrared heater.
| S|N | Year | Field | Application of ANN | R2 Value | Reference |
|---|---|---|---|---|---|
| 1 | 2012 | Food | Prediction of “cold spot temperature profile” of food (using starch dispersion) in retort processing. | 0.9936–0.9955 | (Llave et al., 2012) |
| 2 | 2015 | Food (Cowpea/Legume) | Evaluate the impact of cowpea that has root mucilage on the dissipation of phenanthrene and the composition of microbial community. | - | (Sun et al., 2015) |
| 3 | 2016 | Food (Wheat/Grain) | Prediction of the hydration properties of wheat | 0.994 | (Shafaei et al., 2016) |
| 4 | 2016 | Food (Cowpea/Legume) | Selecting cowpea genotypes with phenotypic stability, together with high stability, aimed at improving recommendations to farmers | - | (Barroso et al., 2016) |
| 5 | 2020 | Food (Cowpea/Legume) | Identification of different varieties of cowpea seeds within large storage facilities via their electrical features | 0.96 | (Audu et al., 2020) |
| 6 | 2020 | Food (Legume)/Stored Products | Evaluating and predicting the density of bruchids within legumes that are bulk stored | 0.89–0.99 | (Banga et al., 2020) |
| 7 | 2021 | Food/Agriculture | Developing a model for predicting rice yield based on transplanting parameters | 0.994 | (Basir et al., 2021) |
| 8 | 2021 | Food/Acoustics | Predicting the likelihood of enjoying food with the existence of background noise | 0.70 | (Alamir, 2021) |
| 9 | 2021 | Food/Gasification | Evaluation of parameters affecting the constituent and yield of syngas from the gasification of food wastes | 0.9813–0.9942 | (Shenbagaraj et al., 2021) |
| 10 | 2021 | Food (Legume) | Predicting the green chickpea’s hydration kinetics at varying temperature of soaking | > 0.9 | (Kumar et al., 2021) |
| 11 | 2022 | Food/Pharmaceutical | Predicting the effect of food on the bioavailability of drugs | 0.82-Training 0.72-Testing | (Bennett-Lenane et al., 2022) |
2.1. Materials
Cowpea (Vigna unguiculata: Agrinawa variety) seeds were received from the South African Agricultural Research Council's Institute for Tropical and Subtropical Crops in Nelspruit, South Africa. The seeds were manually selected to remove faulty seeds and kept at 4°C until moisture preconditioning, and infrared heat treatment were done.
2.2. Experimental design and Sample preparation
Using Statistica version 7 statistical software, a series of experiments were statistically constructed based on RSM-central composite design (CCD) (StatSoft, Tulsa, USA). Level of moisture, infrared heating temperature, and time were the independent variables of pre-treatments investigated, with intervals of 32–57%, 114–185°C, and 2–18 min, respectively. The selection of the parameter levels was based on other studies in the literature on the production of infrared heated cowpea seeds [3, 8, 11, 12]. Bulk density, WABS, and pectin solubility were the dependent variables of pretreatments assessed in this study. The combination of variables resulted in the creation of fifteen (15) experimental runs, each of which was carried out in triplicate, resulting in the generation of 45 rows of experimental data (Table 3). The cowpea seeds were first soaked in water to attain 32, 40, 45, 54, and 57% moisture (dry basis) using the method reported by Mwangwela, Waniska [11] and infrared heated using a closed system infrared heater (MW184, Delphius Technologies, Pretoria, South Africa) for each experimental run (Table 3). The samples were maintained at 4°C in an airtight container until further analysis.
2.2.2. Analyses
Bulk density of cowpea seed samples was determined using the method described by Alves, Da Silva [24]. Water absorption capacity of cowpea seed samples was determined using the method described by Ogundele and Emmambux [25]. Soluble pectin, hot water-soluble pectin (HWSP) of cowpea seed samples was determined as described by Ndungu, Emmambux [12]. The experimental data obtained from the analyses are presented in Table 3.
2.3. Artificial neural network (ANN) modeling
ANN is a supervised form of machine learning driven by the availability of dataset, where raw dataset is the input into the neural system used in developing a predictive model, and subsequently a set of predicted outputs. The neural system consists of hidden layers of neurons that are useful in learning the patterns specific to the raw data and assists in producing possible related outputs [26, 27]. A simple representation of the neural network structure used in our paper is presented in Fig. 1. Differences between the output value in the raw data and the new set of output produced after subjecting our data to ANN algorithm results in an error [27].
2.4. Data Cleaning
The raw data was cleaned and utilized as the input for machine learning prediction using ANN. The predictive variables are moisture temperature and time, while the response variables are bulk density, WABS, and pectin solubility (Table 2). The descriptive property of the raw data is presented in Table 2, while the raw data used as input is presented in Table 3.
| Predictors | Actual Responses | |||||
|---|---|---|---|---|---|---|
| Properties | Moisture (%) | Temp (⁰C) | Time (min) | Bulk-density (g/kg) | WABS (%) | Pectin Solubility (%) |
| Minimum | 32.65 | 114.72 | 2.00 | 0.59 | 95.01 | 146.79 |
| Maximum | 57.34 | 185.27 | 18.58 | 0.67 | 128.95 | 253.23 |
| Median | 45.00 | 150.00 | 8.00 | 0.62 | 108.67 | 180.64 |
| Mean | 46.06 | 150.00 | 8.70 | 0.63 | 111.46 | 189.02 |
| Standard Deviation | 6.96 | 19.69 | 5.17 | 0.02 | 12.09 | 27.29 |
| Standard Error | 1.03 | 2.93 | 0.77 | 0.00 | 1.80 | 4.06 |
| The summary in Table 2. is for the 45 numbers of experiments (15*3replicates) | ||||||
| Moisture | Temperature | Time | Bulk density | WABS | Pectin Solubility | |
|---|---|---|---|---|---|---|
| S/N | (%) | (⁰C) | (min) | (g/kg) | (%) | (%) |
| 1 | 40.00 | 130.00 | 2.00 | 0.66 | 127.63 | 180.75 |
| 2 | 40.00 | 130.00 | 2.00 | 0.66 | 125.18 | 177.42 |
| 3 | 40.00 | 130.00 | 2.00 | 0.66 | 123.81 | 179.73 |
| 4 | 40.00 | 130.00 | 14.00 | 0.64 | 115.60 | 167.11 |
| 5 | 40.00 | 130.00 | 14.00 | 0.64 | 114.38 | 164.07 |
| 6 | 40.00 | 130.00 | 14.00 | 0.64 | 113.98 | 165.08 |
| 7 | 40.00 | 170.00 | 2.00 | 0.65 | 128.71 | 161.34 |
| 8 | 40.00 | 170.00 | 2.00 | 0.65 | 125.31 | 164.38 |
| 9 | 40.00 | 170.00 | 2.00 | 0.66 | 126.45 | 163.14 |
| 10 | 40.00 | 170.00 | 14.00 | 0.62 | 97.50 | 190.76 |
| 11 | 40.00 | 170.00 | 14.00 | 0.62 | 95.96 | 190.15 |
| 12 | 40.00 | 170.00 | 14.00 | 0.61 | 96.00 | 190.47 |
| 13 | 54.00 | 130.00 | 2.00 | 0.62 | 128.95 | 170.74 |
| 14 | 54.00 | 130.00 | 2.00 | 0.61 | 126.22 | 171.65 |
| 15 | 54.00 | 130.00 | 2.00 | 0.62 | 125.14 | 170.52 |
| 16 | 54.00 | 130.00 | 14.00 | 0.63 | 98.40 | 221.09 |
| 17 | 54.00 | 130.00 | 14.00 | 0.63 | 100.39 | 224.12 |
| 18 | 54.00 | 130.00 | 14.00 | 0.63 | 101.09 | 223.22 |
| 19 | 54.00 | 170.00 | 2.00 | 0.61 | 126.06 | 178.02 |
| 20 | 54.00 | 170.00 | 2.00 | 0.61 | 126.00 | 182.27 |
| 21 | 54.00 | 170.00 | 2.00 | 0.61 | 125.49 | 180.89 |
| 22 | 54.00 | 170.00 | 14.00 | 0.59 | 103.97 | 208.35 |
| 23 | 54.00 | 170.00 | 14.00 | 0.59 | 101.09 | 213.20 |
| 24 | 54.00 | 170.00 | 14.00 | 0.60 | 102.11 | 211.73 |
| 25 | 45.00 | 150.00 | 8.00 | 0.62 | 104.16 | 227.76 |
| 26 | 45.00 | 150.00 | 8.00 | 0.61 | 102.00 | 229.58 |
| 27 | 45.00 | 150.00 | 8.00 | 0.62 | 103.96 | 228.59 |
| 28 | 32.65 | 150.00 | 8.00 | 0.67 | 116.66 | 174.08 |
| 29 | 32.65 | 150.00 | 8.00 | 0.66 | 118.16 | 197.13 |
| 30 | 32.65 | 150.00 | 8.00 | 0.66 | 119.36 | 187.14 |
| 31 | 57.34 | 150.00 | 8.00 | 0.61 | 95.01 | 172.56 |
| 32 | 57.34 | 150.00 | 8.00 | 0.60 | 95.31 | 173.78 |
| 33 | 57.34 | 150.00 | 8.00 | 0.61 | 96.00 | 173.91 |
| 34 | 45.00 | 114.72 | 8.00 | 0.64 | 127.08 | 146.79 |
| 35 | 45.00 | 114.72 | 8.00 | 0.63 | 124.67 | 148.91 |
| 36 | 45.00 | 114.72 | 8.00 | 0.64 | 125.10 | 147.65 |
| 37 | 45.00 | 185.27 | 8.00 | 0.60 | 102.78 | 248.99 |
| 38 | 45.00 | 185.27 | 8.00 | 0.59 | 104.65 | 253.23 |
| 39 | 45.00 | 185.27 | 8.00 | 0.60 | 104.07 | 249.88 |
| 40 | 45.00 | 150.00 | 18.58 | 0.61 | 98.01 | 179.54 |
| 41 | 45.00 | 150.00 | 18.58 | 0.61 | 99.60 | 193.49 |
| 42 | 45.00 | 150.00 | 18.58 | 0.61 | 98.02 | 180.22 |
| 43 | 45.00 | 150.00 | 8.00 | 0.62 | 111.46 | 179.24 |
| 44 | 45.00 | 150.00 | 8.00 | 0.62 | 105.67 | 182.57 |
| 45 | 45.00 | 150.00 | 8.00 | 0.62 | 108.67 | 180.64 |
2.5. Data Processing
The processing of data using ANN includes data training, data validation, and data testing. For this research work, MATLAB software was used for ANN modeling. The algorithm used for data processing is the Levenberg-Marquardt algorithm, owing to its swiftness and steadiness in convergence [28]. For data training, the raw data is given as input to the neural network system. During this stage, it fine-tunes the data in retrospect to the error produced. The validation process estimates network’s generalization, besides, signifies when the training process should stop once there is no more progress in data generalization, lastly, data testing evaluates and offers a stand-alone measure of the network’s performance before, during and after data training. For data training, validation, and testing, the dataset was rationed in 70, 15, and 15% respectively. 10 hidden neurons were used for deriving the best predictive model for bulk density and WABS, while three neurons (5,8, and 10) were used for pectin solubility (Fig. 1). This was to produce improved predictive model for pectin solubility.
2.6. Performance of Predictive Model
Understanding the performance of a model is key to improve prediction accuracy. The performance indicators used to evaluate the accuracy of the predictive model developed in this work includes mean square error (MSE), coefficient of correlation (R) and coefficient of determination (R2). Besides, the performance of the ANN model developed was compared with the performance of conventional models such as linear, two factor interaction (2FI), quadratic and cubic model using R2. This was to affirm the accuracy of utilizing ANN for developing a predictive model for estimating the bulk density, WABS, and pectin solubility of cowpea precooked using infrared heating.
A predictive model was developed using Artificial Neural Network for the bulk density, WABS and pectin solubility of cowpea precooked using infrared heating. Three main factors that influence the response were fed into the neural network as the independent variables, the experimental data of the process was cleaned and processed using fitnet ANN model. The MSE, R and R2 of the training, validation, and testing datasets are presented in (Table 4). However, the overall R and R2 value was utilized in selecting the best predictive model for each of the response variable. Besides, scatter plots with coefficient of correlation are presented in (Fig. 2).
The plot of the best validation performance, together with the actual, predicted response and error (difference between the predicted and actual response) plot are presented in Fig. 3 and Fig. 4 respectively. From the evaluation result, for the bulk density, the overall R and R2 values are greater than 0.9, implying that ANN was efficient in developing a predictive model for the bulk density of cowpea prepared using infrared heating.
Specifically, the highest overall R and R2 values are 9.8740E-01 and 9.7496E-01 respectively with a validation MSE of 1.06E-05. This was compared with the R2 obtained using linear, 2FI and quadratic model (Table 5). Comparatively, the R2 of the predictive model generated using ANN was approximately higher by 21%, 13%, and 3% for linear, 2FI and quadratic model respectively.
For WABS, the overall R and R2 ranged between 0.5 and 0.9991. After repeated training, validation and testing, a significantly high R and R2 values were eventually obtained at the 5th iteration with R and R2 values of 9.9911E-01 and 9.9821E-01 respectively with a validation MSE of 3.75E + 00. Similar to the bulk density, the R2 value was compared with that obtained using linear, 2FI and quadratic model (Table 5). Comparatively, the R2 of the predictive model generated using ANN was approximately higher by 28%, 23%, and 11% for linear, 2FI and quadratic model respectively.
For Pectin solubility, the coefficient of regression for the linear, 2FI and quadratic model for pectin solubility are very low, ranging from 0.3–0.49, however, using ANN, a high R and R2 was obtained after training, validating, and testing the model using three different neurons (5,8 and 10 neurons). The three different neurons employed for pectin solubility were utilized to obtain a significant overall model performance indicator (R and R2). The best overall R and R2 values were obtained using 10 neurons and at the 5th run, with values of 0.938 and 0.88 respectively, and validation MSE of 245. Comparatively the R2 value of the model predicted using ANN was approximately higher by 64%, 54%, and 44% for linear, 2FI and quadratic model respectively. The predictive model developed using ANN are presented in equation.
The R2 value from this study is within the range to those reported as indicated in Table 1. The results of comparison between the actual and predicted values as shown in Fig. 5, indicated that the values of the parameters measured in this study (Bulk density, WABS and pectin solubility) are closely related. Reportedly, ANN has been utilized in various fields, nonetheless reports on using ANN for predicting the cooking properties of cowpea when moisture content, temperature and time are varied during infrared heating was not found. This study fills that gap by presenting a method of using Artificial intelligence technologies (specifically artificial neural network) for predicting the bulk density, WABS and pectin solubility of cowpea precooked via the use of infrared form of heating under varied independent parameters of moisture content, temperature, and time.
Table 4. MSE, R and R2 Values obtained from Model Training, Validation and Testing.
<
|
Bulk Density |
|||||||||||||||
|
|
1st RUN |
2ND RUN |
3RD RUN |
4TH RUN |
5TH RUN |
||||||||||
|
|
MSE |
R |
R2 |
MSE |
R |
R2 |
MSE |
R |
R2 |
MSE |
R |
R2 |
MSE |
R |
R2 |
|
Training |
8.77E-06 |
9.9048E-01 |
9.8105E-01 |
9.87E-06 |
9.8995E-01 |
9.8000E-01 |
5.58E-06 |
9.9193E-01 |
9.8392E-01 |
5.36E-06 |
9.9387E-01 |
9.8777E-01 |
2.35E-05 |
9.7855E-01 |
9.5756E-01 |
|
Validation |
1.29E-05 |
9.9035E-01 |
9.8080E-01 |
1.06E-05 |
9.8114E-01 |
9.6263E-01 |
5.32E-05 |
9.7521E-01 |
9.5104E-01 |
6.76E-05 |
9.5105E-01 |
9.0449E-01 |
1.49E-05 |
9.8897E-01 |
9.7807E-01 |
|
Testing |
3.66E-05 |
9.5684E-01 |
9.1555E-01 |
2.38E-05 |
9.7778E-01 |
9.5606E-01 |
3.87E-05 |
9.6548E-01 |
9.3215E-01 |
4.93E-05 |
8.1903E-01 |
6.7082E-01 |
5.08E-05 |
9.7085E-01 |
9.4256E-01 |
|
Overall |
|
9.8573E-01 |
9.7166E-01 |
|
9.8740E-01 |
9.7496E-01 |
|
9.8133E-01 |
9.6301E-01 |
|
9.6262E-01 |
9.2664E-01 |
|
9.7802E-01 |
9.5652E-01 |
|
WABS |
|||||||||||||||
|
|
1st RUN |
2ND RUN |
3RD RUN |
4TH RUN |
5TH RUN |
||||||||||
|
|
MSE |
R |
R2 |
MSE |
R |
R2 |
MSE |
R |
R2 |
MSE |
R |
R2 |
MSE |
R |
R2 |
|
Training |
4.18E+01 |
8.3982E-01 |
7.0529E-01 |
1.67E+00 |
9.9395E-01 |
9.8793E-01 |
1.64E+00 |
9.9415E-01 |
9.8834E-01 |
2.53E+00 |
9.9024E-01 |
9.8057E-01 |
2.34E+00 |
9.9173E-01 |
9.8352E-01 |
|
Validation |
1.03E+02 |
9.0581E-01 |
8.2048E-01 |
9.79E+00 |
9.7925E-01 |
9.5892E-01 |
6.37E+00 |
9.7162E-01 |
9.4405E-01 |
6.55E+00 |
9.8165E-01 |
9.6363E-01 |
3.75E+00 |
9.8802E-01 |
9.7619E-01 |
|
Testing |
2.51E+02 |
7.2171E-01 |
5.2086E-01 |
3.24E+00 |
9.9270E-01 |
9.8545E-01 |
3.91E+00 |
9.9089E-01 |
9.8187E-01 |
7.43E+00 |
9.8121E-01 |
9.6276E-01 |
2.67E+00 |
9.9437E-01 |
9.8876E-01 |
|
Overall |
|
7.5224E-01 |
5.6587E-01 |
|
9.8927E-01 |
9.7866E-01 |
|
9.9057E-01 |
9.8123E-01 |
|
9.8668E-01 |
9.7354E-01 |
|
9.9911E-01 |
9.9821E-01 |
|
Pectin Solubility (70 15 15 5 neurons) |
|||||||||||||||
|
|
1st RUN |
2ND RUN |
3RD RUN |
4TH RUN |
5TH RUN |
||||||||||
|
|
MSE |
R |
R2 |
MSE |
R |
R2 |
MSE |
R |
R2 |
MSE |
R |
R2 |
MSE |
R |
R2 |
|
Training |
1.08E+02 |
9.0191E-01 |
8.1344E-01 |
8.65E+01 |
9.3759E-01 |
8.7907E-01 |
6.19E+01 |
9.6242E-01 |
9.2625E-01 |
1.81E+02 |
8.7271E-01 |
7.6162E-01 |
8.45E+01 |
9.4470E-01 |
8.9245E-01 |
|
Validation |
9.94E+00 |
9.9988E-01 |
9.9976E-01 |
1.42E+01 |
9.9102E-01 |
9.8212E-01 |
2.74E+02 |
9.3218E-01 |
8.6896E-01 |
5.38E+01 |
8.7556E-01 |
7.6661E-01 |
8.11E+01 |
9.5924E-01 |
9.2013E-01 |
|
Testing |
1.34E+02 |
9.1965E-01 |
8.4575E-01 |
1.98E+02 |
8.1991E-01 |
6.7225E-01 |
6.70E+01 |
6.7995E-01 |
4.6233E-01 |
1.97E+02 |
9.8189E-01 |
9.6410E-01 |
1.19E+02 |
8.8885E-01 |
7.9006E-01 |
|
Overall |
|
9.3126E-01 |
8.6725E-01 |
|
9.3451E-01 |
8.7331E-01 |
|
9.33E-01 |
8.6993E-01 |
|
9.0123E-01 |
8.1222E-01 |
|
9.3685E-01 |
8.7769E-01 |
|
Pectin Solubility (70 15 15 8 neurons) |
|||||||||||||||
|
|
1st RUN |
2ND RUN |
3RD RUN |
4TH RUN |
5TH RUN |
||||||||||
|
|
MSE |
R |
R2 |
MSE |
R |
R2 |
MSE |
R |
R2 |
MSE |
R |
R2 |
MSE |
R |
R2 |
|
Training |
1.32E+02 |
9.0341E-01 |
8.1615E-01 |
1.16E+02 |
9.1034E-01 |
8.2872E-01 |
8.25E+01 |
9.3505E-01 |
8.7431E-01 |
9.54E+01 |
9.2787E-01 |
8.6095E-01 |
1.14E+02 |
9.3029E-01 |
8.6544E-01 |
|
Validation |
7.40E+01 |
9.5608E-01 |
9.1409E-01 |
1.41E+01 |
9.9875E-01 |
9.9750E-01 |
1.00E+02 |
9.6600E-01 |
9.3315E-01 |
1.38E+02 |
9.4704E-01 |
8.9688E-01 |
1.79E+02 |
9.1097E-01 |
8.2986E-01 |
|
Testing |
2.16E+02 |
8.6983E-01 |
7.5660E-01 |
4.83E+01 |
7.3384E-01 |
5.3851E-01 |
9.97E+01 |
9.3803E-01 |
8.7990E-01 |
2.92E+01 |
9.7816E-01 |
9.5680E-01 |
5.32E+01 |
9.5776E-01 |
9.1731E-01 |
|
Overall |
|
9.0392E-01 |
8.1707E-01 |
|
9.3699E-01 |
8.7795E-01 |
|
9.3786E-01 |
8.7958E-01 |
|
9.3547E-01 |
8.7510E-01 |
|
9.1876E-01 |
8.4412E-01 |
|
Pectin Solubility (70 15 15 10 neurons) |
|||||||||||||||
|
|
1st RUN |
2ND RUN |
3RD RUN |
4TH RUN |
5TH RUN |
||||||||||
|
|
MSE |
R |
R2 |
MSE |
R |
R2 |
MSE |
R |
R2 |
MSE |
R |
R2 |
MSE |
R |
R2 |
|
Training |
6.21E+01 |
9.5473E-01 |
9.1150E-01 |
1.53E+02 |
8.8425E-01 |
7.8190E-01 |
5.17E+01 |
9.5468E-01 |
9.1141E-01 |
1.04E+02 |
9.2294E-01 |
8.5182E-01 |
4.79E+01 |
9.6429E-01 |
9.2985E-01 |
|
Validation |
2.04E+02 |
8.4422E-01 |
7.1270E-01 |
1.28E+02 |
9.1687E-01 |
8.4065E-01 |
1.75E+02 |
9.2084E-01 |
8.4794E-01 |
1.01E+01 |
9.5736E-01 |
9.1653E-01 |
2.45E+02 |
8.9312E-01 |
7.9767E-01 |
|
Testing |
2.16E+02 |
9.0902E-01 |
8.2631E-01 |
3.27E+02 |
8.5306E-01 |
7.2772E-01 |
1.71E+02 |
7.8957E-01 |
6.2343E-01 |
1.05E+02 |
9.6088E-01 |
9.2329E-01 |
1.07E+02 |
8.3117E-01 |
6.9084E-01 |
|
Overall |
|
9.2629E-01 |
8.5801E-01 |
|
8.7398E-01 |
7.6384E-01 |
|
9.3689E-01 |
8.7776E-01 |
|
9.3695E-01 |
8.7788E-01 |
|
9.3804E-01 |
8.7992E-01 |
|
Response |
Bulk density |
WABS |
Pectin solubility |
|---|---|---|---|
|
Variables |
M,T,t |
M,T,t |
M,T,t |
|
Predictive Models |
Present Work |
Present Work |
Present Work |
|
Linear[R2] |
0.7700 |
0.7143 |
0.3140 |
|
2FI[R2] |
0.8511 |
0.7653 |
0.4034 |
|
Quadratic[R2] |
0.9477 |
0.8864 |
0.4904 |
|
ANN [R] |
0.9874 |
0.9991 |
0.9380 |
|
ANN[R2] |
0.9750 |
0.9982 |
0.8777 |
| M: Moisture, T: Temperature, t: time | |||
Industry 4.0 epitomizes the fourth wave of industrial revolution and aims at applying computerized, artificial intelligent, and data-driven technologies to research and industrial operations in order to optimize productivity and efficiency. One of these technologies is artificial neural network that works on the principle of self-learning to develop accurate predictive models that can describe and predict the relationship and behavior of a system. For the first time, artificial neural network was utilized in this work to develop a model that predicts the behavior (bulk density, WABS and pectin solubility) of cowpea prepared using infrared heating). Unlike the traditional linear, 2FI and quadratic model, Artificial Neural Network (ANN) proved more accurate and developed a better predictive model in the model prediction with a highly significant model performance (R value of 0.9874, 0.9991 and 0.93804 for bulk density, water absorption capacity and pectin solubility) of infrared-cooked cowpea. The predictive model generated can predict similar response variables subjected to similar or almost-close process parameters (moisture content, temperature and time) and paves the path for the optimization of the characteristics of infrared-cooked cowpea using Artificial Intelligence technologies.
- Horn, L.N. Shimelis, H., Production constraints and breeding approaches for cowpea improvement for drought prone agro-ecologies in Sub-Saharan Africa. Annals of Agricultural Sciences, 2020.
- ElMasry, G., Mandour, N., Ejeez, Y., Demilly, D., Al-Rejaie, S., Verdier, J., Belin, E. Rousseau, D., Multichannel imaging for monitoring chemical composition and germination capacity of cowpea (Vigna unguiculata) seeds during development and maturation. The Crop Journal, 2021.
- Kayitesi, E., Duodu, K.G., Minnaar, A. de Kock, H.L., Effect of micronisation of pre‐conditioned cowpeas on cooking time and sensory properties of cooked cowpeas. Journal of the Science of Food and Agriculture, 2013. 93(4): p. 838-845.
- Otunola, G.A. Afolayan, A.J., Evaluation of the physicochemical, proximate, and sensory properties of moinmoin from blends of cowpea and water yam flour. Food science & nutrition, 2018. 6(4): p. 991-997.
- Fabbri, A.D. Crosby, G.A., A review of the impact of preparation and cooking on the nutritional quality of vegetables and legumes. International Journal of Gastronomy and Food Science, 2016. 3: p. 2-11.
- Wang, N., Hatcher, D., Tyler, R., Toews, R. Gawalko, E., Effect of cooking on the composition of beans (Phaseolus vulgaris L.) and chickpeas (Cicer arietinum L.). Food Research International, 2010. 43(2): p. 589-594.
- Wang, N., Hatcher, D.W., Warkentin, T.D. Toews, R., Effect of cultivar and environment on physicochemical and cooking characteristics of field pea (Pisum sativum). Food Chemistry, 2010. 118(1): p. 109-115.
- Ogundele, O.M., Gbashi, S., Oyeyinka, S.A., Kayitesi, E. Adebo, O.A., Optimization of Infrared Heating Conditions for Precooked Cowpea Production Using Response Surface Methodology. Molecules, 2021. 26(20): p. 6137.
- Ogundele, O.M. Kayitesi, E., Influence of infrared heating processing technology on the cooking characteristics and functionality of African legumes: a review. Journal of food science and technology, 2019. 56(4): p. 1669-1682.
- Aboud, S.A., Altemimi, A.B., RS Al-HiIphy, A., Yi-Chen, L. Cacciola, F., A comprehensive review on infrared heating applications in food processing. Molecules, 2019. 24(22): p. 4125.
- Mwangwela, A.M., Waniska, R.D. Minnaar, A., Hydrothermal treatments of two cowpea (Vigna unguiculata L. Walp) varieties: effect of micronisation on physicochemical and structural characteristics. Journal of the Science of Food and Agriculture, 2006. 86(1): p. 35-45.
- Ndungu, K.E., Emmambux, M.N. Minnaar, A., Micronisation and hot air roasting of cowpeas as pretreatments to control the development of hard‐to‐cook phenomenon. Journal of the Science of Food and Agriculture, 2012. 92(6): p. 1194-1200.
- Oyeyinka, S.A., Oyedeji, A.B., Ogundele, O.M., Adebo, O.A., Njobeh, P.B. Kayitesi, E., Infrared heating under optimized conditions enhanced the pasting and swelling behaviour of cowpea starch. International Journal of Biological Macromolecules, 2021. 184: p. 678-688.
- Singh Sibian, M. Singh Riar, C., Optimization and evaluation of composite flour cookies prepared from germinated triticale, kidney bean, and chickpea. Journal of Food Processing and Preservation, 2021. 45(1): p. e14996.
- Zhu, H.-G., Wang, Y., Cheng, Y.-Q., Li, Z.-G. Tong, L.-T., Optimization of the powder state to enhance the enrichment of functional mung bean protein concentrates obtained by dry separation. Powder Technology, 2020. 373: p. 681-688.
- Félix-Medina, J.V., Montes-Ávila, J., Reyes-Moreno, C., Perales-Sánchez, J.X.K., Gómez-Favela, M.A., Aguilar-Palazuelos, E. Gutiérrez-Dorado, R., Second-generation snacks with high nutritional and antioxidant value produced by an optimized extrusion process from corn/common bean flours mixtures. LWT, 2020. 124: p. 109172.
- Soltani, S., Shojaei, T.R., Khanian, N., Choong, T.S.Y., Rashid, U., Nehdi, I.A. Yusoff, R.B., The implementation of artificial neural networks for the multivariable optimization of mesoporous NiO nanocrystalline: biodiesel application. RSC Advances, 2020. 10(22): p. 13302-13315.
- Torrecilla, J., Otero, L. Sanz, P., A neural network approach for thermal/pressure food processing. Journal of Food Engineering, 2004. 62(1): p. 89-95.
- Farkas, I., Remenyi, P. Biro, A., A neural network topology for modelling grain drying. Computers and Electronics in Agriculture, 2000. 26(2): p. 147-158.
- Farkas, I., Remenyi, P. Biro, A., Modelling aspects of grain drying with a neural network. Computers and Electronics in Agriculture, 2000. 29(1-2): p. 99-113.
- Afaghi, M., Ramaswamy, H. Prasher, S., Thermal process calculations using artificial neural network models. Food Research International, 2001. 34(1): p. 55-65.
- Junior, S.B., Mastelini, S.M., Barbon, A.P.A., Barbin, D.F., Calvini, R., Lopes, J.F. Ulrici, A., Multi-target prediction of wheat flour quality parameters with near infrared spectroscopy. Information processing in agriculture, 2020. 7(2): p. 342-354.
- Mutlu, A.C., Boyaci, I.H., Genis, H.E., Ozturk, R., Basaran-Akgul, N., Sanal, T. Evlice, A.K., Prediction of wheat quality parameters using near-infrared spectroscopy and artificial neural networks. European food research and technology, 2011. 233(2): p. 267-274.
- Alves, J.L.F., Da Silva, J.C.G., Di Domenico, M., Galdino, W.V.D.A., Andersen, S.L.F., Alves, R.F. De Sena, R.F., Exploring Açaí seed (Euterpe oleracea) pyrolysis using multi-component kinetics and thermodynamics assessment towards its bioenergy potential. BioEnergy Research, 2021. 14(1): p. 209-225.
- Ogundele, O.M. Emmambux, M.N., Effect of infrared heating of pre-soaked whole and dehulled bambara groundnut (Vigna subterranea) seeds on their cooking characteristics and microstructure. LWT, 2018. 97: p. 581-587.
- Dawood, T., Elwakil, E., Novoa, H.M. Delgado, J.F.G., Toward urban sustainability and clean potable water: Prediction of water quality via artificial neural networks. Journal of Cleaner Production, 2021. 291: p. 125266.
- Finke, A., Escobar, J., Munoz, J. Petit, M., Prediction of Salt Spray Test results of Micro Arc Oxidation coatings on AA2024 alloys by combination of Accelerated Electrochemical Test and Artificial Neural Network. Surface and Coatings Technology, 2021: p. 127370.
- Yu, H. Wilamowski, B.M., Levenberg–marquardt training, in Intelligent systems. 2018, CRC Press. p. 12-1-12-16.
No competing interests reported.