Application of artificial neural network for prediction of fluoride removal efficiency using neutralized activated red mud from aqueous medium in a continuous fixed bed column

The present research work approaches the removal of fluoride from aqueous medium using neutralized activated red mud (NARM) in a continuous fixed bed column. Artificial neural network (ANN) technique was applied effectively for optimization of the model for the practicability of the removal process. The consequences of various experimental variables, like bed length, adsorbate concentration, experimental time, and adsorbate solution flow rate are studied to know the breakthrough point and saturation times. The highest removal potentiality of NARM was considered to be 3.815 mg g−1 of F− in the bed height of 15 cm, starting concentration 1 ppm, susceptible time 120 min, adsorbate solution flow rate 0.5 mL min−1, and constant room temperature, respectively. Bohart-Adams and Thomas models were considered to describe the fixed bed column effect to the bed height and adsorbate concentrations. The experimental data were applied to a back propagation (BP) learning algorithm programme with a four-seven-one architecture model. The artificial neural network model was considered to be functioning correctly as absolute relative percentage error throughout the learning period. Differentiation between the predicted outcomes from ANN model and actual results from experimental analysis affords a high degree of correlation (R2 = 0.998) stipulating that the model was able to predict the adsorption efficiency. Experimented adsorbent materials were characterized using different instrumental analysis that is scanning electron microscopy–energy dispersive X-ray spectroscopy (SEM–EDS), Fourier transform infrared spectroscopy (FTIR), and X-ray diffraction (XRD).


Introduction
Population detonation, anomalous climate conditions, development of modern civilization, metropolitan city development, rapidly progressing industries, and advanced technology have generated enormous amounts of toxicants to the severe contamination both in ground and surface water bodies (Wang et al. 2018;Chakraborty and Das 202). The water pollution arises due to the different environmental conditions that brings consequential consideration of its remediation skills and techniques (Mukherjee and Halder 2018;Nguyen et al. 2013). Current situation in India fluoride infected in ground water has been ascertained in more than 20 states and 66 million people are in distress from fluorosis disease (Maitra et al. 2020;Bhattacharya 2016). As a result of higher concentration and longtime contamination of fluoride ions reveals a lot of problems formed on the human health. Accordingly, the World Health Organization (WHO) has suggested the standard concentration of 1.5 mgL −1 F − in drinking water purposes (Wan et al. 2021;Ali et al. 2016). This limit has recommended a challenge for present research of new technologies capable of selectively removing low levels of F − . In contemplation of fluoride analysis in water, developed a cost-efficient innovation to bring fluoride from aqueous medium to great consciousness at the end of 20 years. Various physico-chemical processes like deterioration, precipitation-coagulation, biodegradation and absorption have been practiced to remove fluoride from water (Mohapatra et al. 2009; Khandare and Mukherjee 2019;Deshmukh et al. 2009;Sivasankar et al. 2010;Daifullah et al. 2007). Traditionally various conventional and non-conventional techniques have been enforced to eliminated F − from contaminated water bodies. Adsorbent surface assimilation technique is an adsorption process of best application of adsorbate ions are completely separated out from aqueous medium. It is a very simple technique, economically very less, cost-efficient, and environmentally friendly nature (Camargo 2003;Gandhi et al. 2012;Chen et al. 2021;Angelin et al. 2021;Alagumuthu and Ranjan 2010a). Red mud is an eccentric waste material and possessions of different high amounts of toxic heavy metals that is silicon dioxide [SiO 2 (3-50%)], aluminum oxide [Al 2 O 3 (10-20%)], ferric oxide [Fe 2 O 3 (30-60%)], calcium oxide [CaO (2-8%)], sodium oxide [Na 2 O (2-10%)], and titanium dioxide [TiO 2 (trace-10%)] as well as assemblage of minor ingredients likely potassium, chromium, vanadium, nickel, barium, copper, manganese lead, and zinc (Sahu et al. 2010). Consequently, development and carrying out its storage, remediation programs remain essential, and its inventory grows approximately 120 million per annum. The red mud can be simply transported from the dumping site to the nearby residential areas. Contact of the red mud causes serious threat to the ground water by seepage and also causes various environmental hazards (Tor et al. 2009;He et al. 2013;Bhatnagar et al. 2011). Neutralized activated red mud prepared from fresh red mud, and its use for the fixed bed column adsorption of the F − from aqueous medium has not been reported. The highest F − removal efficiency using fixed bed columns from aqueous medium of different activated adsorbents are presented in Table 1. The adsorption treatment process is a complicated method because of different parameters are interconnection with the removal process. Artificial neural network (ANN) is a powerful prediction tool for the identification of the relevant parameters and their correlation relationships are very complex. Removal of toxicants from aqueous solutions using different prediction models has been used currently by number of researchers (Chen et al. 2021;Wan et al. 2021). The main objectives of the current research works were using a multilayer feed forward neural network model to forecast the removal of F − from aqueous medium using fixed bed column study. The input parameters encompass adsorbate concentration, fixed bed height, experimental time, and adsorbate flow rate whereas removal of F − is considered as output results. The prophecy of breakthrough curve point is calculated from different fixed bed column length and adsorbate concentration using Bohart-Adams and Thomas equations. The total experimental results can be separated into learning and inference sets, if the network system is well trained, the ANN model can be tested with any data to optimize the best outcome. To know the adhesion of F − exterior sit of the NARM using different instrumental characterization that is SEM-EDS, FTIR, and XRD.

Preparation of adsorbent and reagents
The collected fresh red mud samples were first grinded in motor pestle and the powder was sieved using a 450 mm sieving machine. Then, 20 gm of fresh red mud powder was taken a 250-ml washed container and added 100 ml deionized water to it and mixed properly. The counterbalanced form of fresh red mud by slowly put in 1 N nitric acid (HNO 3 ) and 1 N sodium hydroxide (NaOH) fresh prepared solutions. The pH all the experimental solutions was determined by using a Systronics pH meter. Fifty grams of neutralized red mud was taken in a 250-ml washed dry beaker and 70 ml of concentrated 98% H 2 SO 4 was added slowly, after 10 min of reaction, 30 ml of concentrated 65% HNO 3 was added in to the sample. Consequently, the treated samples were kept 24 h in a well-maintained oven, after that the prepared samples were washed four times carefully using deionized water further put in the oven at 120 ºC for completely dried. Figure 1 manifests a graphically encapsulated detailed preparation procedure of activated neutralized red mud from fresh red mud. The physico-chemical parameters of the neutralized activated red mud results are particle size (100-240 µm), pH (7), moisture contained (8.01%), conductivity (21.13 µS cm −1 ), specific gravity (0.31), porosity (81%), bulk density (0.55 gmL −1 ), ion-exchange capacity (0.79 meqg −1 ), water soluble matter 1.16%), and volatile matter (41%), respectively. The specific surface area of neutralized activated red mud is found to be 212.02 m 2 g −1 (Brunauer-Emmett-Teller surface area analyzer, Quantachrome AUTOSORB-1, USA). All experimental analysis, standard preparation and chemical reagent preparation using fresh collected deionized water. One thousand part per million of fluoride stock solutions was processed by adding 0.221 g of NaF solutions in a 1-L of deionized water, this stock solutions having 10 ppm of F − . Subsequently, to prepare a sequence of standard fluoride solution, desirably 0.5 ppm F − by dilution of the stock solutions (Clesceri et al. 1989;Miretzky and Cirelli 2011).

Fixed bed column studies
Fluoride removal from contaminated water in a fixed bed column experiments by neutralized activated red muds were carried out using standard 0.1 ppm, 1 ppm, 5 ppm, and 10 ppm fluoride solution in the absence of other competing ions. Column studies were accomplished in a glass column of 6 cm internal diameter, 50 cm bed height, and samples flow rate are asserted by a burette, respectively ( Fig. 1). In the bottom portion of the glass column was placed a purified cotton to the prevent any loss of sample materials and mechanical support to the filtrate bed. Total experiment was carried out at room temperature. Effects of process parameters like fixed bed column height (adsorbent dosage) was assorted as 5 cm (6.50 g), 10 cm (12 g), and 15 cm (18.5 g), fluoride concentration is assorted as 1 ppm, 5 ppm, and 10 ppm, experimented duration was assorted as 10-150 min and flow rate of sample was assorted as 0.5 mLmin −1 , 2.5 mL min −1 , and 5.5 mL min −1 were investigated. Experimental samples were collected from the bottom of the glass column and tested to know the F − concentration in the samples. After treatment each sample was allowed to settle for 10 min and it is centrifuged at 1000 rpm for 20 min and filtered through Whatmann 42 µ size filter paper. The concentrations of the F − before and after adsorption were determined by Orion star A214 Fluoride electrode (Thermo Scientific). The continuous fixed bed column potentiality was evaluated by the breakthrough time period and removal quantity given in Eq. 1 (Chen et al. 2021;Tor et al. 2009).
To know the effective efforts of the fixed bed glass column systems applying various easy numerical models were successful. In the present, research work has been approaching two simple mathematical models that are Bohart-Adams and Thomas. Bohart-Adams model indicated that the rate of the removal process is corresponding to the part of removal efficiency that persists on the sorbent materials. Bohart-Adams model can be intimated in Eqs. 2 and 3 (Chen et al. 2021; Ye where C 0 is the initial adsorbate concentration in ppm, C t is the removal of adsorbate ions in ppm with respect to experimental time period in minutes, x B is the mass of adsorbate ions removed in the fixed bed column at breakthrough (mg), m is the mass of the NARM in the column (g), Qv is the flow rate (mLmin −1 ), C B is the breakthrough adsorbate ions concentration (mgg −1 ), t B is the experimental time to breakthrough (min), K AB illustrated the Adams-Bohart kinetic constant (mgL −1 ), N 0 is the saturation concentration(mgL −1 ), t is the sample flow time (min), Z stands for fixed bed (1) , and U is the linear flow rate (ml. min −1 ), S stands for bed cross section area (cm 2 ) and X stand for unit mass of adsorbent packed in the glass column (g). K th is the Thomas rate constant (ml.mg −1 min −1 ), q 0 is the maximum capacity of adsorption (mgg −1 ), X quantity of the adsorbent in the column (g). V ef is volume of solution (ml) and F is the flow rate (ml.min −1 ).

Back propagation neural network model (BPNN)
The adsorbate ions removal effectiveness calculated using different analytical appliances are intricated because-of complication of the system. So, the artificial neural network system is embraced to enhance the forecast ability of removal efficiency at contact time period. Accordingly, in the present research results implemented in ANN modelling for optimization purposes as a consequence of high potentiality of distinguished the input variables and output results in the convoluted condition. The back propagation neural network (BPNN) system comprises of three distinct layers (Banan et al. 2020;Fan et al. 2020). The first input layer (P) collects data from experimental analysis then proceeds this data into the network system for process of treating. The second hidden layer (I) obtained data from the input layer, after that all the collected data treating in the network system. Finally output layer (Y) accepts clarified data from the network system, then it conveys the outcomes to an outside effector. The shape of the back propagation network system manifested in Fig. 2. The neural network functioning system consists of two stages that is first learning or training stage and second inference or testing stage. The arrangement of this network system as it may be conferred E-I-Y, the input layer (E) stipulated by the number of input parameters. The input indicators are reform along associated with significance is known as performance factor (Pmn), which illustrates the connected of n th neuron of the input layer to mth neuron of the hidden layer. Similarly, the output indicators of hidden layer are reform by connected with performance factor (Pom) of Oth neuron of output layer to mth node of the hidden layer. All the reform instruction used a logistic sigmoid transfer factor (f) are incorporated at the output layer (Liu et al. 2018;Nikzad et al. 2015;Wang et al. 2016). Let I s = (I S1, I S2 , I S3 ………. I Si ), S = 1, 2, 3……. E is S th design among E input manners. Where Pom and Pmn are the relationship importance between nth input neuron to mth hidden neuron, and mth hidden neuron to Oth output neuron, respectively Dashti et al. 2021). The estimated output results from a neuron in the input, hidden, and output layer using a mathematical equation described as Sigmoid transfer function (f) is a circumscribed, sameness, continuance. S-shaped function that provide a classified nonlinear response. It includes the logistic sigmoid function: The linear transfer function is used as the output layer transfer function (f 0 ). In this work multilayered forward ANN with seven hidden layer was used. For all data sets, sigmoid transfer function in the hidden layer and a linear transfer function in the output node were used. All experimental analysis results are accomplished by MATLAB 7 using a Pentium IV PC mathematical software with the ANN toolbox (Saha et al. 2010). In present studies, the four input parameters (adsorbate solutions, fixed bed column height, experimental time, and adsorbate flow rate) are utilized in an artificial neural network software for desired output efficiency of the adsorbent by column studies. Presently 170 data are produced from laboratory analysis and it is separated into two sets that is learning and inference used for the modeling system. ANN network processing system 75% data are used for the training set and 25% data are used for testing set. All the data are conciliated in the 0.1-0.9 range to elude the measuring consequences of parameter values. Consequently, each and every data (X i ) are changed to normalized results (X norm ) as obeys and Demirel, 2008;Aleboyeh et al. 2008): where X i is the input or output variable X, X min , and X max are the minimum and maximum value of variable X. At first, the combination weights are caused to arbitrarily in the range of − 1 to + 1. Fixed bed column type of operated learning studies has been embraced in the present situation, whereas correlation weights are accommodated using delta rule algorithms after propelled the whole training data to the network system. By utilizing the actual and predicted conclusion to investigate the precision of an artificial network system (Yetilmezsoy and Demirel 2008;Wang et al. 2018;Chawdhury and Saha 2013). The finest evidence efforts of artificial network model achieve by utilizing the root mean square error (RMSE) described as.
where q e model is the predicted result and q e experimental is the actual value for the ith manner. Weight change at any time t, is given by

Consequences of fixed bed column length with experimental time on the breakthrough curve
The bed length is the key variable for the removal of fluoride from aqueous medium in the continuous fixed column studies. The breakthrough curve illustrates the ratio between initial concentration of fluorides /remaining concentration of fluoride in the experimental container (C t /C) against the adsorbate solution flow rate time (t). Figure 3 described as the continuous fixed bed column removal process of fluoride at different bed heights 5 cm (Fig. 3a), 10 cm (Fig. 3b), and 15 cm (Fig. 3c), experimental time (10-150 min), at constant starting concentration 1 ppm, flow rate 0.5 ml min −1 , room temperature, and at neutral pH, respectively. Table 2 illustrates the mathematical results of fixed bed columns with different bed height and experimental time of fluoride solutions. The bed height is strongly influenced by the fixed bed column breakthrough saturation time and the adsorbent bed length performance. Mathematical results of breakthrough curve and saturation times increase with increase in length of bed height and exposed time of fluoride solution because of developed more numbers of attaching sites on the upper surface of the adsorbent during the removal processes (Saha et al. 2010;Yetilmezsoy and Demirel 2008). The present studies shows that the breakthrough time increased with the increases in bed length. The maximum removal of fluoride is considered to be 3.815 mg g −1 , bed height 15 cm, fluoride concentration 1 ppm, flow rate 0.5 ml min −1 , at room temperature, respectively. When pH of the aqueous medium remains in a neutral range (pH = 7) the fluoride removal onto the neutral adsorbent surface can be decreased by a ligand or ion exchange reaction mechanism (Chen et al. 2021;Ranasinghe et al. 2022;Ghorai and Pant 2005). Figure 3 manifested a differentiation between the ANN model predictions and the experimental data as a function of length of bed height. From this, figure clearly shows that the ANN model satisfactorily predicts the trend of the experimental data.

Consequences of adsorbate concentration with experimental time on the breakthrough curve
Adsorbate concentrations is the major variable for the influence of the removal effectiveness of fluoride ions in  the continuous fixed bed column system. Figure 4 delineation the continuous fixed bed column removal process of fluoride ions at different fluoride concentrations 1 ppm (Fig. 4a), 5 ppm (Fig. 4b), and 10 ppm (Fig. 4c), experimental time (10-150 min), a constant length of bed height 15 cm, flow rate 0.5 mL/min, neutral pH, and room temperature, respectively. At the maximum removal of fluoride ions shown in 1 ppm because the surface of neutralized activated red mud bed saturated quickly leading to breakthrough and exhaust in time. The higher concentration of fluoride ions needs to disseminates on the exterior surface of the adsorbents by intraparticle dissemination processes. So, the adsorbate to form hydrolysed ions will spread out slowly. Some scientists have published that the primary adsorbate concentrations provide a driving force to conquer mass and convey resistance of F − ions between the aqueous medium and adsorbent surface. This designated the possible monolayer formation of F − ions on the outer surface of the neutralized activated red mud (Chawdhury and Saha 2013;Amirkhani et al. 2021). Experimental results showed that the removal process reached equilibrium in about 120 min of flow rate for all concentrations of fluoride ions in this study. However, for initial fluoride ion concentrations of 1, 5, and 10 ppm, adsorption equilibrium percentages are determined to be higher than the fluoride concentration of 5 ppm. Hence, 1 ppm of fluoride ions concentration was selected as the optimal starting concentration for further breakthrough experiments. Figure 4 represented the experimental results with comparison of ANN output results for different starting concentrations.

Consequences of adsorbate solutions flow rate with experimental time on the breakthrough curve
Flow rate is one of the most significant characteristics in evaluating the removal capability of the adsorbent for continuous fixed bed column treatment of fluoride containing aqueous solutions. The consequence of flow rate in the fixed bed column system with neutralized activated red mud is examined changing flow rate to 0.5 mL/min (Fig. 5a), 2.5 mL/min (Fig. 5b), and 5.5 mL/min (Fig. 5c) with constant fluoride concentration 1 ppm, neutral pH 7, bed height 15 cm, and room temperature, respectively. The highest removal of fluoride ions from aqueous medium shown in flow rate of 0.5 ml min −1 because of the residence time of the adsorbate was more at lower flow rate and so the adsorbent got more time for binding with fluoride ions effectively. Subsequently permitting F − ions to ingress more alive sites within the adsorbent. For the higher flow rate 2.5 ml.min −1 and 5.5 ml min −1 the breakthrough curve (C t /C 0 ) of fluoride is established to increase at the initial part of the treatment but at the meantime flow rate time is increased in the operation time shows that decrease in removal ability. Increasing the adsorbate solution flowrate shows the inclination increase of the breakthrough curve. Consequently, increasing the adsorbate solution flow rate the external film diffusion mass transfer resistance decreases, concluding in fast saturation and untimely breakthrough time. Furthermore, decreases in adsorbate solution flow rate, the intra-particle diffusion becomes more effective due to longer residence time. The speed of the initial concentration of the fluoride ions remarkably affected the contact between the surface of the adsorbate and adsorbent (Chawdhury and Saha 2013;Joshi et al. 2020;Deshmukh et al. 2009). The experimental data and ANN calculated outputs for various flow rate values are shown in Fig. 5.

Consequences of experimental time on the breakthrough curve
Experimental time is the most compelling feature in assessing the adsorption of F − ions from water using neutralized The consequence of experimental time in the continuous fixed bed column process with adsorbents is treated at different contact experimental time 10 min to 150 min with constant fluoride concentration 1 ppm, neutral pH 7, bed height 15 cm, flow rate 0.5 ml min −1 and room temperature, respectively. The results are represented as percentage removal of F − versus experimental time. The percentage adsorption of F − with initial adsorbate concentration 1 ppm, increased from 70.26 to 89. 13% (Fig. 4a) for 10-150 min experimentations. It is evident from the results that more than 90% adsorption occur within the first 60 min and equilibrium is attained after 90 min. The development of the rate of adsorption as it may be primarily all adsorbent sites are empty and the adsorbate concentration slope is high. So, the intermolecular attraction between supportive energy sites on the adsorbent and independently electronic properties of the molecules involved in this reaction. The fluoride adhesion on the surface of the neutralized activated red mud by weak van der Waals forces and multiple layers may be formed at constant heat. Figure 4a represented the experimental results with comparison of ANN output results for different experimental times.

Breakthrough curve analysis by Bohart-Adams and Thomas model
Bohart-Adams and Thomas models are appliance to forecast the continuous fixed bed column achievement variables. The consequences of breakthrough curves prognosticate analysis by the Bohart-Adams model are differentiated among the investigational result with different bed length and constant starting concentration of fluoride 10 ppm and outcomes are conferred in Fig. 6a. This model prognosticates the different variables observed from experimental data function as enhanced the fixed bed column performance. Table 3 illustrated the various parameters results of this model are interpenetrate adsorbate concentration (N 0 ), Bohart-Adams rate constant (K AB ), adsorbate solution flow rate, and fluoride concentration, correlation coefficient, and root mean square error, respectively (Chen et al. 2021;Chawdhury and Saha 2013). It is considerably from Fig. 6a and Table 3 exhibits an impecunious concurrence between actual result and prognosticated results. In the current investigation, the interpenetrate adsorbate concentration (N 0 ) results gradually diminished among the enlargement of fixed bed length because of greater extent area are void for removal of fluoride ions. Concluded that the removal process is bestowal by the corporeal mass convey of the breakthrough curve analysis (Geleta et al. 2022;Ghorai and Pant 2005). Rate constant (K AB ) of this model is raised among the enlargement in the bed length disclosed the kinetics constant is conquered by extraneous mass convey in the beginning portion of the fixed bed column system. To conclude that, Bohart-Adams model handled a clear, understandable and extensive perspective for evaluating the removal of fluoride ions from aqueous medium. Figure 6b indicated the consequences of breakthrough curves prognosticate analysis by the Thomas model are compared with the actual results. Different variables are estimated using Thomas model equation that is rate constant (K TH ) of Thomas model, maximal solid-phase concentration (q 0 ), adsorbate solution flow rate, fluoride concentration, correlation coefficient, root mean square error and the values are exhibited in Table 3. The diagrammatic and enumeration values exhibit a satisfactory concurrence among actual and prognosticate consequence suggests the Thomas model (Deshmukh et al. 2009;Zare et al. 2022).
In the present study, Thomas rate constant (K TH ) results reveals optimum at the moment concentration of fluoride is minimal denoted the fixed bed column removal process of the adsorbents is kinetically agreeable at minimal ranges of contamination. The solid phase concentration (q 0 ) is reduced among the rise in the fixed bed length as a consequence of the rate limiting step being moved from peripheral to inside mass conveyance. To concluded that, Thomas model presumed adsorption arise at particular uniform area inside of the materials and Pseudo-second order kinetics activities is suitable for break through curve analysis (Mohan et al. 2017;Sahu et al. 2020).

Prediction approach for removal efficiency using ANN model
The prophecy of removal of efficacious fluoride ions from water utilizing activated neutralized red mud (ANRM) is a complicated hypothesis, consequently neural network prediction system approach is espoused. Total 168 investigational results assembled from laboratory observations after that are separated into two sets:75% (126) data are applied to the learning phase and 25% (42) data are applied to the inference phase. Back propagation neural network probabilistic program along three layered based architectonic structure narrated aside arched convey assignment at input layer, one hidden layer is operated, and straight away convey function is manipulated in an output layer, respectively and it is presented in Fig. 2 (Rojas et al. 2015). Primarily, the learning results are furnished in to the neural network systems and its dissemination of output results (Fig. 7). The number of neurons used in the hidden layer is manifested by the connection of two into whole root over of the number of variables used in the input system plus one. In the network system to observe the best number of neurons at the hidden layer by determining the mean squared error (MSE) (Yetilmezsoy and Demirel 2008;Chawdhury and Sahu 2013). In present study, four number of variables are selected in the input layer, seven neurons are chosen in the hidden layer since MSE commences diminish, and finally predicted results represented in one output layer. Training and impetus variables are put down at range 0.25 and 0.20, respectively (Wang et al. 2016;Joshi et al. 2020;Zhang et al. 2022). In the learning phase initially responsibility as the input results is transmitted ahead across the network system to enumerated the output result of each unit. The estimation of error indication for each output unit in the neural network system bank on predicted output results contrast with desired results.
Root means square error (RSME) is 0.371 calculated from a neural network system dependent upon the number of exhaustive that overtake the learning results. Figure 8a represented as correlation coefficient (R 2 = 0.998) for learning set result observed from actual and predicted removal percentage of fluoride ions. In the course of the inference phase the dissemination of output results and an extremity of correlation coefficient (R 2 = 0.987) between actual and predicted removal percentage of fluoride ions is achieved as shown in Fig. 8b. The mean absolute relative percentage error is considered to be 0.671  for learning and 0.640 for inference data set generated from a neural network processing system. Juxtaposition results of the actual and predicted for inference data sets are shown in Table 4. The application of artificial neural network model is a best attempt towards enhancing the prediction coherence of adsorbate ions from aqueous medium. Even so, the present research work is consciousness on patterning of fluoride ions removal from water using a new ecofriendly adsorbent media by continuous fixed bed column method (Liu et al. 2018;Raji et al. 2022). In general, the investigational learning point of view functions in the best effort of artificial neural network application in designing of integrated wastewater analysis.  Instrumental description of neutralized activated red mud German) images are obtained at 50.00 K x magnification with corresponding energy dispersive spectrum of the activated neutralized red mud (ANRM) prior to and afterward fluoride ions adsorption at neutral pH and it is illustrated in Fig. 9. Fresh neutralized activated red mud without approach of fluoride analysis distinctly displays the surface consistency, architecture and different range of apertures (Fig. 9a). Afterward neutralized activated red mud treated with 1 ppm fluoride ions evidently changes in the upper surface. It is distinctly noticed that the upper surface of the treated adsorbents has been changed into a novel glossy bit and crystals are embedded in the surface (Fig. 9b). The corresponding EDS spectra of neutralized activated red mud control and treated with 1 ppm of fluoride ions is shown in Fig. 9a FTIR (Perkin Elmer FT-IR, RX-I) studies are done in order to know the functional groups and structure of the neutralized activated red mud. Figure 10 delineation the spectra of neutralized activated red mud prior to and afterward removal of fluoride ions exhibits number of absorption peaks. The spectra of prior to and afterward removal of fluoride ions is illustrated in Fig. 10a, b. In the present FTIR analysis discloses that some absorption bands shifted from higher to lower frequency may be stretching or bending vibration with different functional groups. Fluoride ions feasibly complication to the numbers of absorption peaks that is 3,305.99 cm -1 is moved to 3,251.98 cm −1 because Fig. 9 SEM-EDS analysis of a before and b after treatment of F − 1 ppm using NARM of cemented with − OH groups, 1, 631.78 cm −1 has been moved to1,600.92 cm −1 because of connected with N-H and C = O stretching vibration, 1,481.33 cm −1 has been moved to 1,419.61 cm −1 as a consequence of connected with N-H groups, and 887.28 cm −1 moved to 883.40 cm −1 because of corresponding to the O-C-O scissoring vibration of polysaccharide, respectively (Angelin et al. 2021;Rojas et al. 2015;Ye et al. 2018;Chen et al. 2021;Gandhi et al. 2012;Sivasankar et al. 2010;Prabh and Meenakshi 2014). Finally, concluded that the transfer of the spectra from higher frequency to lower frequency probably ascribed to the interconnection of F − among the hydroxyl, amide and amino groups afford on the exterior surface of the neutralized activated red mud. XRD (PHILLIPS X'PERT X-ray diffractometer (model PW 1710) image of the neutralized activated red mud control and treated with 1 ppm fluoride solution is shown in Fig. 11. Fresh neutralized activated red mud generally composed of different elements like O, Si, Al, Fe, and Ti are exhibited in EDS image (Fig. 9a). The XRD data of the treated materials contributed evidence of decreases in the peak intensity which shows the adsorption of fluoride ions in comparison with control materials. So, it was concluded that part of fluoride ions is converted into HF, some part of the converted into AlF 3 and another some parts converted into FeF 2 at neutral pH and finally fluoride ions adhesion on the surface of the neutralized activated red mud (Zare et al. 2022;Chen et al. 2021).

Conclusions
The present research work successfully demonstrates that a continuous fixed bed column technique and artificial neural network mathematical tool is effective for the removal and prediction of fluoride from water using neutralized activated. Prediction efficiency is observed using backpropagation algorithms in three layered based neural network model. For learning and inference data set applied in neural network processing system generated mean absolute relative percentage error is 0.671 and 0.640, respectively. High degree correlation coefficient (R 2 = 0.995) is the learning data set and conveys 0.37 root means square error estimated from the neural network system. Behavior of breakthrough modelling and equilibrium times exhibits the optimum removal ability is considered to be 3.815 mgg −1 . The better prophesy of breakthrough curve analysis is assessed by Bohart-Adams and Thomas model. The treated samples are analyzed by different instrumental techniques. FTIR spectra may be attributed to the interaction of fluoride ions with the amide, amino and hydroxyl groups present on the exterior surface of the neutralized activated red mud. XRD pattern manifests part of fluoride ions converted into HF, AlF 3 , and FeF 2 . So, neutralized activated red mud is the best choice of adsorbent media for fluoride ions removal from aqueous medium. Artificial neural network is a statistical prediction tool with high-speed confluent conceivably among investigational results to minimize analytical endeavor. Continuous fixed bed column adsorption studies have faced some limitations like difficult to maintain temperature, undesirable heat gradients formed, undesirable chemical reactions, conveying, and column washed problems. These studies are thought about future improvements and challenges for fixed bed column adsorption process as feed adsorbate solutions is continuously moved towards the selected column, mass transfer zone gradually move through the fixed bed, investigated of after particle time exposed, the adsorbent particles inlet or outlet of the mass transfer zone do not participate in the mass-transfer processes, adsorbing the adsorbate ions