Abstract
Wastewater is a serious concern for the environment. There is a substantial amount of toxins that are discharged continuously from several pharmacological companies that lead to serious damage to public health and the ecosystem. Present wastewater treatment technologies include primary, tertiary, and secondary treatments that remove numerous contaminants; but pollutants in the nanoscale range were hard to remove with these steps. Some of these include inorganic and organic pollutants, pathogens, pharmaceuticals, and pollutants of developing concern. The utility of nanoparticles was a promising solution to this issue. Nanoparticles have exclusive properties permitting them to potentially eliminate residual pollutants but being eco-friendly and inexpensive. This study develops a new Archimedes optimization algorithm (AOA) with Stacked Sparse Denoising Auto-Encoder (SSDAE) model, named AOA-SSDAE for wastewater management in the IoT environment. The presented AOA-SSDAE technique aims to predict wastewater treatment depending on the influent indicators. In the presented AOA-SSDAE technique, the IoT devices are initially employed for the data collection process and then data normalization is performed to transform the collected data into a uniform format. For the predictive process, the SSDAE model is employed in this paper. To improve the SSDAE model's prediction capability, the AOA-based hyperparameter tuning process is involved.
HIGHLIGHT
AOA-SSDAE is a novel Archimedes optimization algorithm (AOA) model with Stacked Sparse Denoising Auto-Encoder (SSDAE) for wastewater management in the context of the Internet of Things (IoT).
The AOA-SSDAE method presented here aims to forecast wastewater treatment based on indicators of the influent.
In the provided AOA-SSDAE methodology, IoT devices are initially used for data collection, followed by data normalization to convert the collected data into a standard format.
INTRODUCTION
Climate change and the fast-growing world population are the two main factors that massively affect freshwater accessibility. Infections brought on by dangerous organisms present in water result in 19% of fatalities worldwide, whereas poor water quality causes 80% of illnesses (Wang et al. 2022). The threat to the state of the universe as it was evolving was made worse by the difficulty and expense of analysing the chemical and microbial makeup of water. Besides harming human life, polluted water even damages the environment and wildlife (Epelle et al. 2022). In the last 20 years, globally, the water pipes connected to residences have augmented, but still, a large group of people have no access to clean drinking water, nearly 780 million (Kamali et al. 2019). The water conservation technique is utilized for reducing water usage, but more new solutions are required as conservation is not sufficient because of the needs of the increasing population. Poor sanitation even hinders the supply of safe and clean water (Nizamuddin et al. 2019). Hence, new cost-effective and innovative ways to treat wastewater are vital. The fundamental wastewater treatment process generally includes three main steps: primary, tertiary, and secondary treatment (Kumar 2021).
Negative environmental issues like solid waste management, air pollution, and water pollution are being resolved using nanotechnology (Dutta et al. 2022). The more complicated impurities to remove in wastewater lie in a nanoscale range of 1–100 nm; therefore, the nano-based methods are the most suitable techniques. Nanotechnology is very beneficial for water remediation not due to the dimensional field, but then due to its outstanding physicochemical property of nanomaterials (Manimegalai et al. 2022). Activated carbon refers to adsorption characterized and its structure of porous, thermostability utilized in numerous applications like the purification of wastewater treatment, removal of impurities, gaseous phases, and odor from medical usage. Using this activated carbon (Guduru et al. 2021), adsorption removes the wastewater colours fast and affordably. The treatment of wastewater is complex. Yet, developments in intellectual methods allow them to be used in complex modeling mechanisms (Dang et al. 2022). Owing to their robustness, potential applications in engineering and great precision can be used for the improved provision of performance features. Certain indispensable variables can be utilized for assessing the performance of wastewater treatment plants (Xin et al. 2021). Such factors are total suspended substances (TSS). TSS are water particles larger than 2 microns. A totally dissolved solid is any particle under 2 mm (TDS). TSS includes algae and bacteria, but most of it is inorganic. TSS includes sand, silt, and plankton. Decomposing plant and animal waste in water sources releases suspended solids. Some silt settles at the bottom of a water source, but other TSS floats or remains suspended. TSS affects water quality. Chemical oxygen demand (COD) is the amount of dissolved oxygen that must be present in water for chemical organic molecules like petroleum to be oxidized. The COD concentration in receiving waters is evaluated to determine the short-term impact of wastewater effluents on the oxygen levels in such waters, and biological oxygen demand (BOD). Biochemical oxygen demand is the dissolved oxygen (DO) aerobic biological organisms need to break down organic material in a water sample at a certain temperature and time. The BOD, calculated as the quantity of oxygen needed per litre of sample after 5 days of incubation at 20 °C, is used to calculate the level of organic contamination in water. Successful wastewater treatment facilities reduce BOD. Wastewater effluent BOD can determine the immediate influence on receiving water oxygen levels. The COD test is less specific because it tests all chemically oxidized organic compounds. Such features were utilized as a model for wastewater treatment plants.
This study develops a new Archimedes optimization algorithm (AOA) with Stacked Sparse Denoising Auto-Encoder (SSDAE) model, named AOA-SSDAE for wastewater management in the IoT environment. The presented AOA-SSDAE technique aims to predict wastewater treatment depending on the influent indicators. In the presented AOA-SSDAE technique, the IoT devices are initially employed for the data collection process and then data normalization is performed for transforming the collected data into a uniform format. Normalization creates clean data. However, data normalization has two purposes: it makes all records and fields look the same. It improves entry type coherence, cleansing, lead creation, segmentation, and data quality. For logical data storage, this procedure eliminates unstructured data and duplicates. Data normalization properly standardizes data entry. The BOD, measured as the amount of oxygen required per litre of the sample after 5 days at 20 °C incubation, is used to quantify the level of organic contamination in water (Gangathimmappa et al. 2022). Benefits of data normalization: More storage, more rapidly answered inquiries, improved segmentation, the data must be normalized; there is no other option. For the predictive process, the SSDAE model is applied in this study. For improving the predictive performance of the SSDAE approach, the AOA-based hyperparameter tuning process is involved. A widespread experimental results investigation is conducted to make sure of the improvements of the AOA-SSDAE algorithm.
RELATED WORKS
Saka et al. (2022) focused on using sunlight for catalyzing the destruction of carbon-based (organic) pollutants. To improve the proficiency of the photocatalytic technique and increase the morphological area, sodium alginate was arranged as a drop practice and utilized as a polymeric tool. SiO2 nanoparticles have been doped as sodium alginate droplets. The proposed method was capable of spreading the wavelength diversity all over the considerable wavelength constituency. The sunlight catalytic process has been carried out from a photo light droplet or UV–Vis. Soffian et al. (2022) investigated a variety of carbon-based materials (CBMs), including biochar, activated carbon, carbon aerogel, graphene, and carbon nanotubes. The generation of dissimilar CBMs and their alteration receives special consideration. As detrimental dyes, organic, and inorganic pollutants evolving from the wastewater have given rise to damage to the water supplies and environment. Adsorption is the more commonly used technique for the deduction of harmful pollutants owing to its usability and comparatively low cost compared to other techniques.
Madhura et al. (2019) focused on modern treatment technique with nano-based materials, for instance, carbon and graphene nanotubes, metal nanoparticles including iron, silver, magnesium, and zinc, and magnetic-core composites and metal oxides nanoparticles such as iron, cobalt, and nickel. Moreover, the author compares conventional and emerging techniques for the performance and production cost of organic and inorganic pollutants. Moyo et al. (2022) proposed a summary of the dissimilar kinds of nanocellulose utilized in the manufacturing of nanofiber membranes, their properties, and production routes. Ong et al. (2018) focused on the recent advancements of carbon-oriented nanocomposite membranes and nanomaterials for the potential treatment of water mixtures or emulsified oil. They also investigated influence of bubble size on the effectiveness of the separation process and the effect that this has on the amount of material removed.
In Dhiman et al. (2022), numerous methods for manufacturing carbon-related single-atom catalysts were defined. Following that, an overview of recent advances in a variety of specialized procedures and computer achievements is offered in order to comprehend the geometric and electrical properties of catalysts attached to a single carbon atom. To minimize aquatic pollution, single-carbon atom photocatalysts were used. As single-atom catalysts on carbon particles, many materials are utilized, for example, carbon nitride, graphene, carbon quantum dots, and other materials (Lakshmanna et al. 2022). Sawdust, a waste product that is harmful to the environment, was used to construct a novel activated carbon synthesis (Abdel-Salam et al. 2020). Magnetite nanoparticles were used to further modify it. Two hydrophilic nanoparticles, silica and bismuth oxide nanomaterials, are included in the nanocomposites Mass Attenuation Coefficient (MAC)/Bi2O3 and MAC/SiO2. Sawdust was microwave-pyrolyzed first, and nanoparticles were added by co-precipitation.
In Xin et al. (2021), Fenton-like nanocomposite catalysts made of carbon are employed in microwave, light, electro, and ultrasound Fenton processes. Chemicals are produced and used as a result of industrialization. Organic contaminants in municipal and industrial wastewater are difficult to degrade. They have a huge influence on society if left untreated. More study is needed to better understand Fenton-like processes, how to use them effectively, and how to combine them with artificial intelligence to remediate refractory organic pollutants. Thus, environmental protection necessitates the creation of Fenton-like catalysts. In Manimegalai et al. (2023), the agricultural, pharmaceutical, electronic, and environmental industries have all experienced rapid advancements in nanotechnology. The most promising nanotechnology products for electronics and drug delivery are carbon-based nanoparticles because of their distinctive characteristics. Conventional and novel micro/nano-pollutants caused global warming and climate change.
THE PROPOSED MODEL
Overall workflow of the AOA-SSDAE system.
Data normalization
and 
were the maximal and minimal values of attribute A. Min–Max normalization mapped a value of A to 
in the range by calculating 
Prediction using the SSDAE model
which represents the hidden space as 
. The encoding functionality parameterized using 
, as follows:
weight matrix. The hidden representation 
is later mapped back to the ‘reconstructed’ vector, 
, in input space
; in such cases, the AE is regarded as having tied weights. The main objective of the training is to learn the parameter 
, and 
was to decrease the average reconstructed error over a set of input vectors, namely, 
indicates the loss function so that cross-entropy 
and 
parameters are enhanced using mini-batch or stochastic gradient descent. The network will learn the identity function since the autoencoder has been modified to produce a denoising autoencoder (DAE). Or, to put it another way, if the amount of data is too great, the autoencoder may simply learn it and provide results that are similar to the input. The autoencoder in this scenario does not perform any useful representation learning or dimensionality reduction. Autoencoders with denoising capabilities can be used to resolve this problem. To do this, they purposefully change the input data, either by introducing noise or by concealing some of the input values (Neelakandan et al. 2022). A DAE is a three-layer neural network which comprises a hidden layer (HL), an input layer, and an output layer. After adding noise to the initial input x, the interference input 
is attained, and inputted feature expression h can be attained using the encoding function 
. Next, h is mapped to the resultant layer using the decoder function 
to achieve the reconstruction data z of the input dataset. The encoder method of DAE is mathematically formulated below:
denotes the coding model parameter; w represents the coding weight matrix, and b indicates the coding offset vector. The decoder procedure is mathematically formulated below:
denotes the decoded model parameter, 
indicates the decoded weight matrix, 
specifies the decoding offset vector, and 
represents the non-linear activation function. The reconstructed outcome z was unable to exactly replicate the input dataset x for the training instance set 
, the cost function of DAE is
is a weight representation, and lowering the weight lowers overfitting. Additionally, a penalty component was added to the cost function to restrict the level of activation of the neurons responsible for the high-dimension sparse feature extraction from the input dataset. The sparse DAE (SDAE) cost function can be expressed as a result as
reflects the control's sparse restriction weight, whereas 
denotes the sparsity parameter and 
denotes the average activation of input dataset equivalent to 
nodes of the HLs. The training of SSDAE adopted the greedy layer-wise training approach that could efficiently mitigate the gradient diffusion phenomenon; i.e., every SDAE is separately trained to attain the optimum weight of the network that is exploited as an SSDAE network weight. Backpropagation is the algorithm for computing the gradient, not how it is employed, although it is sometimes used to refer to the full learning algorithm, including stochastic gradient descent. Backpropagation, a specific case of reverse accumulation, generalizes the gradient calculation in the delta rule, which is the single-layer form of backpropagation. Automatic differentiation generalizes backpropagation. Backpropagation tests for faults from output nodes to input nodes. It improves data mining and machine learning predictions mathematically. Backpropagation calculates derivatives quickly. Then, the network is finetuned by the error backpropagation (BP) technique until the optimum parameter of the networks is derived.
Architecture of SDAE.
Hyperparameter tuning process
The Archimedes Optimization Method is a population-based Pigeon Inspired Optimization Algorithm (PIOA) algorithm that is based on the Archimedes principle, a well-known physical fact. Archimedes' principle concisely explains the rule of buoyancy, which determine the relationship between an item immersed in a fluid and the buoyant force exerted on it. If the object's weight is more than the weight of the displaced fluid, it sinks; otherwise, it floats above the fluid in the issue. Finally, the AOA is applied for the optimal hyperparameter tuning of the SSDAE model. AOA can be a population-based metaheuristic physics approach based on Archimedes' rules of buoyancy in water (Geetha et al. 2022). AOA exploits the population of an object as a candidate to accomplish a certain objective. AOA initiates by defining how to use the initial population and later starts to reiterate the procedure with certain limits. The search for objects includes the generation of accelerations, volumes, and densities. The AOA was a globally optimal technique that was characterized by mathematical modelling as follows (Raghavendra et al. 2022):
, density (
), and acceleration (
) as follows:
objects in the population of Population of N Objects (POP) objects, whereas minimal and maximal limitations of solution space were represented as 
and 
, correspondingly. rand generates random numbers within [0,1] in 
-dimension vector. In these steps, the better population is measured, and better fitting objects with maximum fitness values are selected and allocated 
, and 
iterations are upgraded and are expressed by:
represents the random numeric that is distributed regularly within 
and 
denote the volume and density of better objects recognized up to these points.
. These steps include transfer from exploration to exploiting process. A collision among objects is initiated and, after certain time, the object strives for a state of balance. The 
increases till it attains a constant value and is determined as follows:
indicates the maximum number of iterations. TF increases until it reaches 1 over time. The density decreasing factor, 
, which decreases over time and is determined as follows, is an additional parameter that assists AOA with global-to-local search:
By using iteration, the value of 
until it reaches a predetermined target location, it lowers with time. AOA allows a better balance between exploration and exploitation if these parameters are well-regulated.
Step 4 – Update the normalization and acceleration of an object. The acceleration of object for the iteration 
updating procedure 
is divided into three stages: ‘exploration stage,’ ‘exploitation stage,’ and ‘normalized acceleration stage’.
0.5, it specifies a collision between objects, we select the random material (mr) and upgrade the acceleration of the object for 
iteration as follows:
, and 
represent acceleration, density, and volume, of an object l, correspondingly. 
, and
, correspondingly represent density, volume, and acceleration of the random material. Especially, it is necessary to point out that 
allows exploration in one-third of iteration period time. The exploration and exploitation effectiveness is changed once the value except 0.5 is exploited.
, the subsequent formula must be used for updating the acceleration of the object at the following iteration, 
:
signifies the acceleration of an object and is the optimum value.
and 
, correspondingly. 
indicates the proportion of steps that every object changes. The acceleration value is higher if the 
object can be farther from the global optima. On the other hand, it will be in the exploitation stage, namely, representing the progression of search from the exploration stage to the exploitation stage. The AOA was effective to achieve stability regarding exploitation and exploration.
, the location should be upgraded, where the location of object is adapted for the iteration 
as follows:
indicates the 
object's upgraded location in a population of POP objects. 
indicates the constant that often equals 2. In the exploitation stage (
, the object updates the location as:
denotes the constant equivalent to 6. 
progressively increases with time and directly relates to the transfer operator, which is described by 
. The range of 
is 
, and 
is a constant equivalent to 2. Then, a small proportion is initiated; meanwhile, this causes the step size of random walk to be considerable and the difference between the present and the best locations to be substantial. In the search, this proportion slowly increased to reduce the difference between the actual and the best locations, accomplishing a balance between exploitation and exploration. F shows the item's direction as a flag based on the following expression, which is derived from the object's position:
, and 
denotes the constant equivalent to 0.5.
, and 
In this study, the AOA was used for determining the hyperparameter indulged in the SSDAE method. The Mean Square Error (MSE) is the objective function and can be defined below:
and 
designate the appropriate and attained magnitudes for 
unit from the resultant layer of network in time t, respectively.RESULTS AND DISCUSSION
In this section, the experimental validation of the AOA-SSDAE technique is tested using the test dataset. Table 1 represents the details of parameter setting.
Parameter setting
| Parameters | Minimum | Maximum | Average |
|---|---|---|---|
| T-N, mg/dm3 | 19.8 | 99 | 70.4 |
| T-P, mg/dm3 | 3.8 | 38.6 | 13.55 |
| BOD, mg/dm3 | 40.5 | 792 | 378 |
| COD, mg/dm3 | 169 | 2,520 | 930.1 |
| TSS, mg/dm3 | 82 | 1,145 | 440 |
| Q, mg/dm3 | 26,983 | 66,883 | 38,758 |
| Parameters | Minimum | Maximum | Average |
|---|---|---|---|
| T-N, mg/dm3 | 19.8 | 99 | 70.4 |
| T-P, mg/dm3 | 3.8 | 38.6 | 13.55 |
| BOD, mg/dm3 | 40.5 | 792 | 378 |
| COD, mg/dm3 | 169 | 2,520 | 930.1 |
| TSS, mg/dm3 | 82 | 1,145 | 440 |
| Q, mg/dm3 | 26,983 | 66,883 | 38,758 |
Predictive outcome of the AOA-SSDAE approach with existing systems under varying classes
| Class | BKNN | ELM | BKNN-ELM | AOA-SSDAE |
|---|---|---|---|---|
| Mean absolute error (MAE) | ||||
| T-N | 6.06 | 7.21 | 3.98 | 3.11 |
| T-P | 1.62 | 2.85 | 1.45 | 1.04 |
| BOD | 45.32 | 48.15 | 39.92 | 35.30 |
| COD | 119.21 | 139.26 | 109.42 | 98.87 |
| TSS | 57.98 | 77.93 | 48.12 | 45.38 |
| Average | 46.038 | 55.080 | 40.578 | 36.740 |
| Mean absolute percentage error (MAPE) | ||||
| T-N | 9.74 | 12.41 | 7.56 | 6.89 |
| T-P | 14.07 | 18.85 | 11.76 | 10.76 |
| BOD | 14.22 | 15.04 | 13.23 | 12.35 |
| COD | 12.98 | 16.18 | 10.97 | 8.37 |
| TSS | 16.00 | 26.21 | 12.10 | 11.22 |
| Average | 13.402 | 17.738 | 11.124 | 9.918 |
| Coefficient of correlation (R) | ||||
| T-N | 0.76 | 0.29 | 0.36 | 0.28 |
| T-P | 0.48 | 0.54 | 0.40 | 0.39 |
| BOD | 0.60 | 0.65 | 0.61 | 0.54 |
| COD | 0.48 | 0.40 | 0.34 | 0.36 |
| TSS | 0.58 | 0.59 | 0.43 | 0.44 |
| Average | 0.580 | 0.494 | 0.428 | 0.402 |
| Class | BKNN | ELM | BKNN-ELM | AOA-SSDAE |
|---|---|---|---|---|
| Mean absolute error (MAE) | ||||
| T-N | 6.06 | 7.21 | 3.98 | 3.11 |
| T-P | 1.62 | 2.85 | 1.45 | 1.04 |
| BOD | 45.32 | 48.15 | 39.92 | 35.30 |
| COD | 119.21 | 139.26 | 109.42 | 98.87 |
| TSS | 57.98 | 77.93 | 48.12 | 45.38 |
| Average | 46.038 | 55.080 | 40.578 | 36.740 |
| Mean absolute percentage error (MAPE) | ||||
| T-N | 9.74 | 12.41 | 7.56 | 6.89 |
| T-P | 14.07 | 18.85 | 11.76 | 10.76 |
| BOD | 14.22 | 15.04 | 13.23 | 12.35 |
| COD | 12.98 | 16.18 | 10.97 | 8.37 |
| TSS | 16.00 | 26.21 | 12.10 | 11.22 |
| Average | 13.402 | 17.738 | 11.124 | 9.918 |
| Coefficient of correlation (R) | ||||
| T-N | 0.76 | 0.29 | 0.36 | 0.28 |
| T-P | 0.48 | 0.54 | 0.40 | 0.39 |
| BOD | 0.60 | 0.65 | 0.61 | 0.54 |
| COD | 0.48 | 0.40 | 0.34 | 0.36 |
| TSS | 0.58 | 0.59 | 0.43 | 0.44 |
| Average | 0.580 | 0.494 | 0.428 | 0.402 |
MAE outcome of the AOA-SSDAE approach under varying classes.
MAPE outcome of the AOA-SSDAE approach under varying classes.
COC outcome of the AOA-SSDAE approach under varying classes.
of 97.45%, 
of 97.27%, 
of 97.77%, and 
of 97.57%. Simultaneously, with run-3, the AOA-SSDAE technique reaches 
of 96.70%, 
of 97.20%, 
of 97.20%, and 
of 97.31%. Concurrently, with run-4, the AOA-SSDAE technique achieves 
of 96.49%, 
of 97.27%, 
of 96.26%, and 
of 96.84%. Finally, with run-5, the AOA-SSDAE technique obtains 
of 96.41%, 
of 96.19%, 
of 96.08%, and 
of 96.95%. Overall outcome of the AOA-SSDAE approach with different measures and runs
| No. of runs |  |  |  |  |
|---|---|---|---|---|
| Run-1 | 97.45 | 97.27 | 97.77 | 97.57 |
| Run-2 | 97.01 | 96.78 | 96.45 | 96.14 |
| Run-3 | 96.70 | 97.20 | 97.20 | 97.31 |
| Run-4 | 96.49 | 97.27 | 96.26 | 96.84 |
| Run-5 | 96.41 | 96.19 | 96.08 | 96.95 |
| Average | 96.81 | 96.94 | 96.75 | 96.96 |
| No. of runs | | | | |
|---|---|---|---|---|
| Run-1 | 97.45 | 97.27 | 97.77 | 97.57 |
| Run-2 | 97.01 | 96.78 | 96.45 | 96.14 |
| Run-3 | 96.70 | 97.20 | 97.20 | 97.31 |
| Run-4 | 96.49 | 97.27 | 96.26 | 96.84 |
| Run-5 | 96.41 | 96.19 | 96.08 | 96.95 |
| Average | 96.81 | 96.94 | 96.75 | 96.96 |
Overall outcome of the AOA-SSDAE approach with different runs.
of 96.81%, 
of 96.94%, 
of 96.75%, and 
of 96.96%. 
Average outcome of the AOA-SSDAE approach with different runs.
TACY and VACY outcome of the AOA-SSDAE system.
TLOS and VLOS outcome of the AOA-SSDAE system.
Precision-recall outcome of the AOA-SSDAE approach.
ROC outcome of the AOA-SSDAE approach.
of 79.19%, 
of 90.17%, and 
of 90.14%. In addition, the BKNN model obtains slightly improvised outcomes with 
of 88.14%, 
of 78.16%, and 
of 72.19%. Moreover, the BKNN-ELM model obtains reasonable performance with 
of 96.20%, 
of 94.12%, and 
of 93.19%. However, the AOA-SSDAE technique accomplishes better performance with 
of 96.81%, 
of 96.94%, and 
of 96.75%. These results ensured that the AOA-SSDAE technique accomplished the maximum performance with 
of 96.81%, 
of 96.94%, and 
of 96.75%. Comparative analysis of the AOA-SSDAE system with other methods
| Methods | Accuracy | Precision | Recall |
|---|---|---|---|
| BKNN | 88.14 | 78.16 | 72.19 |
| ELM | 79.19 | 90.17 | 90.14 |
| BKNN-ELM | 96.20 | 94.12 | 93.19 |
| AOA-SSDAE | 96.81 | 96.94 | 96.75 |
| Methods | Accuracy | Precision | Recall |
|---|---|---|---|
| BKNN | 88.14 | 78.16 | 72.19 |
| ELM | 79.19 | 90.17 | 90.14 |
| BKNN-ELM | 96.20 | 94.12 | 93.19 |
| AOA-SSDAE | 96.81 | 96.94 | 96.75 |
Comparative analysis of the AOA-SSDAE system with other methods.
CONCLUSION
In this study, we introduce a novel AOA-SSDAE technique for wastewater management in an IoT setting. The presented AOA-SSDAE technique aims to predict wastewater treatment depending on the influent indicators. In the presented AOA-SSDAE technique, a series of sub-processes are involved, namely data collection, data preprocessing, prediction, and parameter tuning. Primarily, the IoT devices are initially employed for data collection process and then data normalization is performed for transforming the collected data into a uniform format. For the predictive process, the SSDAE model is applied in this research. For the SSDAE model's prediction capability to be improved, the AOA-based hyperparameter tuning process is involved. A widespread experimental results analysis is made to ensure the improvements of the AOA-SSDAE system. The AOA-SSDAE methodology, which has a 96.81% accuracy rate, can provide better results. A detailed comparison study revealed that the AOA-SSDAE technique outperformed other modern models. To ensure that the AOA-SSDAE system will be able to be improved even further in the future, further enhancements will make use of more advanced methods.
ACKNOWLEDGEMENTS
A.A.G. is grateful to the Researchers Supporting Project number (RSP2023R407), King Saud University, Riyadh, Saudi Arabia, for the support.
DATA AVAILABILITY STATEMENT
All relevant data are included in the paper or its Supplementary Information.
CONFLICT OF INTEREST
The authors declare there is no conflict.
