Optimal conditions of paint wastewater coagulation with gastropod shell conchiolin using response surface design and artificial neural network-genetic algorithm

The potential of gastropod shell conchiolin (GSC) (a waste product of the deprotenization stage of chitosan production) as one of the alternatives to chemical coagulants has been explored for treatment of paint industrial wastewater (PW). The accuracy of response surface design (RSD) and the precision of artificial intelligence in predicting and optimizing the process conditions were harnessed in raising experimental design matrix and response optimization, respectively for the bench scale jar test coagulation experiment. PW was characterized using American Public Health Association standard methods. Extraction of conchiolin was done via alkaline extraction method. PW contains 2098 mg/l total suspended solid above discharge limit (1905 mg/l). Fourier transform infrared (FTIR) spectrum of GSC revealed a broad N–H wagging band at 750–650 cm−1 indicating the presence of secondary amine linked to the presence of protein. Turbidity removal from PW via one factor at a time was found to be a function of pH, GSC dosage, temperature and time. Artificial neural network response prediction shows 92% correlation with the RSD experimental result. The optimal conditions obtained via genetic algorithm for the response optimization at the best pH of 4 indicate optimal turbidity removal of 98% at GSC dosage, time and temperature of 4 g, 20 min and 45 °C, respectively.


Introduction
Wastewater management describes a general and comprehensive methodology for treatment, discharge or reuse of contaminated domestic or industrial wastewater (Abd-Elaty et al. 2021). Specifically, industrial wastewater management uses systematic methods in decontamination of toxic contaminants in industrial wastewaters to conform to the general acceptable discharge standard (Abd-Elaty et al. 2021;Subramonian et al. 2015). In chemical industries, paint manufacturing industry is among the industries that generate large volume of wastewater (Zhang et al. 2021). Majorly, the wastewater from paint industry comes from batch wash-off. Paint wastewater contains colloidal particles (Zhang et al. 2021;Menkiti et al. 2018) and is also very rich in calcium (Menkiti et al. 2018). These colloidal particles make the wastewater very turbid and of poor quality (Igbinosa and Okoh 2009;Zhang et al. 2022). Discharge of poor quality wastewater into the watershed poses health risk (Igbinosa and Okoh 2009;Zhang et al. 2022;Okoh et al. 2007;Carlson et al. 2013). Also, disposing paint wastewater on land will contaminate the land by introducing unwanted contaminants that can affect the soil fertility (Carlson et al. 2013). Proper treatment of paint wastewater can be achieved by decontamination of the colloidal particles (Zhang et al. 2021;Carlson et al. 2013;Ejimofor et al. 2021a). Researches have been conducted on particles decontamination from wastewaters using different methods such as adsorption (Augusto et al. 2020;Sharma et al. 2019), reverse osmosis (Sadeddin et al. 2011), sedimentation (Song et al. 2000), filtration (Jones et al. 2001) and ultra-filtration (Liu and Zhou 2004). More recent studies reported high particles removal efficiency through coagulation-flocculation [16.17,18]. Coagulation-flocculation process of wastewater treatment involves destabilization of stable negatively charged suspended particles within the wastewater sample by introducing counter charged (positively charged) coagulants . Effective interactions between the negatively charged particles and the coagulants result in charge neutralization, inter-particle bridging, particles double layer compression and sweep flocculation (Ejimofor et al. 2020a). The combined effects of these stages leave behind clarified wastewater samples and settled sludge.
The use of synthetic chemical coagulants such as alum, poly-aluminum chloride and ferric chloride for coagulation experiment have been successfully carried out by many researchers (Ejimofor et al. 2021c;Bahrodin et al. 2021). However, the disadvantages of chemical coagulants which include; increase in pH of treated wastewater, generation of large volume of sludge, high risk of Alzheimer's infection, high cost of residual sludge management, the cost of the coagulants and toxicity level of the generated sludge (Cheng et al. 2005) have been major limitations to the use of chemical coagulants. Hence, the need for alternative opened up new research interests on natural or bio-coagulants (Cheng et al. 2005;Saravanan et al. 2017). Extracts from shells (snail shell chitosan) (Chi and Cheng 2006), Periwinkle shell extract (Ezemagu et al. 2021a), plant-based coagulants (moringa) (Ndabigengesere and Narasiah 1998), mucuna seed extract (Nwabanne et al. 2018;Ezemagu et al. 2021b) have been successfully used in wastewater treatment as bio-coagulants. More precursors are still under investigation of which this work has been able to harness and study one (gastropod shells from livestock farming).
In livestock farming, production and processing of gastropods generate large quantity of shells as waste products. These agricultural wastes can be processed to chitosan through deproteinization and demineralization (Babayemi 2013). During the deprotenization stage, some loosely attached proteins (conchiolin) are removed (Babayemi 2013). These proteins are disposed as waste products.
Hence, proper utilization of this waste-product of chitosan production has been a gap that needs to be addressed. This work therefore was conducted to close this gap. The potentials of conchiolin as coagulant have been accessed in treatment of paint wastewater. Also, due to the robustness of response surface design (RSD), and the proven optimization efficiency of artificial intelligence (genetic algorithm) (Rao et al. 2007), hybrid method (the use of RSD and ANN-GA) was explored for experimental design and optimization of the process response for reduction of colloidal particles in paint wastewater.

Materials
This work used wastewater generated from production of water-based emulsion paint, concentrated coagulants extracted from gastropod shell (conchiolin) and analytical grade NaOH.

Materials collection
The paint wastewater sample was collected after batch washoff from a paint factory in Onitsha, Anambra State Nigeria in March, 2016. The gastropod shells for the extraction of the natural coagulant were source from farmer's livestock market in Onitsha while the reagents were bought from Bridge head market Onitsha.

Characterization of paint wastewater
The paint wastewater sample was characterized using standard methods. Table 1 shows the parameters and their standard testing methods.

Extraction of conchiolin from gastropod shell floor
The gastropod shell conchiolin (GSC, chito-protein) was extracted from gastropod shells using slight modification of the method described by Babayemi (2013) and Rao et al. (2007). 1500 g of gastropod shells were washed, sun  (2012) dried, crushed and sieved using 0.6 mm sieve to obtain gastropod shell flour (GSF). The sieved flour was deproteinized using 3.5% (w/v) of dilute sodium hydroxide (NaOH) solution (Rao et al. 2007). The GSF and the sodium hydroxide solution were mixed in solid-liquid ratio of 1:10 (w/v) in 500 ml beaker. The mixture was stirred (using magnetic stirrer) for 2 h at a constant temperature of 65 °C. During this stage, there was mild formation of foam that eventually receded after few minutes. After the 2 h stirring time, the mixture was cooled to room temperature. The cooled mixture was filtered with filter sack set-up to separate the liquid extract containing the gastropod shell conchiolin (GSC). The filtrate was allowed to settle for 40 min before decantation. The conchiolin obtained after decanting the clear extraction solution was sun dried for two days, ground and stored in an air tight container for use as coagulant. Figure 1, shows the graphical representation of the process route. Before use, the conchiolin sample was subjected to physiochemical analysis to obtain the proximate composition using the standard method as shown in Table 2.

Instrumental techniques for gastropod shell conchiolin analysis
The instrumental analyses were conducted on the Gastropod shell coagulant (GSC) to ascertain some major qualitative and quantitative properties. The Fourier-transform infrared spectroscopy (FTIR) was conducted using Alpha-P spectrometers (Bruker Optics, Bryanston, South Africa), X-ray Diffraction (XRD) was conducted using Bruker Apex-2 Duo single-crystal X-ray diffractometer, differential scanning calorimetric (DSC) pattern and Thermo-gravimetric analysis (TGA) were obtained using Q200 differential scanning calorimeter and TGA Q500 TA (both from TA instruments, Delaware), respectively. The scanning electron microscopic analysis (SEM) was conducted using ZEISS Merlin FE SEM, Germany.

Coagulation-flocculation test
The coagulation-flocculation studies considered the influence of different process variables (pH, coagulant dosage, time, and temperature) on turbidity reduction via one-factor at a time (OFAT) method. The experimental design matrix was obtained using response surface design, while the   AOAC-920.53 (1995) optimal coagulation-flocculation conditions were established via ANN-GA.

Coagulation-flocculation mechanism of gastropod shell conchiolin
Bio-coagulation mechanism followed the actual mechanism of coagulation-flocculation process. It involves interaction between the negatively charged suspended particles and the positively charged coagulant particles (Babayemi 2013). This interaction follows four different mechanisms from the fast stirring (coagulation stage) stage to the floc sedimentation stage. Introduction of Gastropod shell conchiolin into the wastewater sample brings about double layer formation and compression, which in turns result to charge neutralization via charge-charge Columbic attraction (Rao et al. 2007). This mechanism leads to destabilization of the stable colloidal particles resulting to formation of micro-flocs during the fast stirring stage. Within this period, there are net positive and negative charges from the adsorption of the colloidal particles onto the coagulants creating charged loops and tails for further interactions. These two mechanisms govern the coagulation stage of the process. Further slow stirring stage (flocculation stage) lead to inter-particle bridging and nonselective aggregation of more particles onto the available charged sites. The latter is known as sweep flocculation. Furthermore, particles are enmeshed as the flocs aggregate and gain weight. The flocs settle out by gravity leaving behind a clarified bulk of treated water (Ejimofor et al. 2020a;Chi and Cheng 2006).

Effect of coagulant dosage variation
The initial pH and turbidity of the paint wastewater were measured at room temperature using thermo-orion420A + pH meter and MC-WZS-185 turbidity meter, respectively. 1000 ml of the paint effluent contained in six different 1000 ml beakers (gg-17, 10 cm in diameter) were dosed with 0.5, 1, 2, 3, 4 and 5 g of gastropod shell conchiolin (GSC). The mixtures (paint effluent and GSC) were subjected to rapid mixing at 250 rpm (velocity gradient (G) = 310 s −1 ) for 2 min, followed by slow mixing at 30 rpm (G = 22 s −1 ) for 20 min using magnetic stirrer. Afterward, the treated effluent was allowed to settle for 30 min. During the settling period, 20 ml of the supernatant was pipetted into the curvet of the turbidity meter for turbidity measurement at time intervals of 0, 3, 5, 10, 15, 20, 25 and 30 min and readings were recorded in NTU. The residual turbidities recorded in NTU were converted to mg/l by multiplying NTU by 2.35 as described by Ezemagu et al. (2021a). Where, 2.35 is a factor for converting turbidity (NTU) to mg/l.

Effect of pH variation
The best coagulant dosage obtained from "Effect of coagulant dosage variation" section was used for the evaluation of pH effect. Equal quantities of the already established coagulant dosage were dosed into 6 different beakers with each containing 1000 ml of paint wastewater sample. The pH of the paint wastewater samples were adjusted to 2, 4, 6, 8 and 10 using 0.1 M H 2 SO 4 and 0.1 M NaOH before dosing of the coagulant. 2 min of fast (250 rpm) stirring, followed by 20 min (30 rpm) of slow stirring periods were performed on each of the mixtures. Thereafter, the residual turbidities were measured using turbidity meter within the settling period of 35 min at intervals of 3, 5, 10, 15, 20, 25, 30 and 35 min.

Effect of temperature variation
A set of Jar tests were carried out using the values of dosage and pH already determined in "Effect of coagulant dosage variation" and "Effect of pH variation" sections. The temperature of the solution was adjusted to 25, 35 and 45 °C using magnetic stirrer, fitted with hot plate (0-100 °C) (B.Bran Scientific model). 2 min fast stirring at 250 rpm followed by 20 min slow stirring at 30 rpm were performed on each mixture. The residual turbidity was measured and recorded after the flocculation period, between 0-35 min for every 5 min intervals.

Artificial neural network response prediction and optimization
ANN system was used for response prediction, while Genetic algorithm was used for process optimization. The proposed route for ANN response prediction and GA optimization is illustrated in Fig. 2. Experimental design matrix was raised using central composite design (CCD) on response surface methodology (RSM) environment via design expert 12. Three factors coagulant dosage (1-5 g), time (5-30 min) and temperature (25-45 °C) and a response variable (turbidity removal efficiency) were considered as input and output variables, respectively. The design matrix generated (Table 3) was used for the actual jar test experiment. According to process route in Fig. 2, the result of the experiment were used for ANN network modeling, training, testing and validation producing output variables (predicted responses) according to the run order in Table 1. The predicted responses were compared with the experimental result and optimized using genetic algorithm.
Response prediction model was developed via back propagation neural network (BP) method in ANN-MATLAB 12 environment. BP algorithm constitutes of input, hidden layer and output with neurons as the basic processing unit. Prediction model is developed by connecting the different layers of data with appropriate weights (w) and biases (b) (Rao et al. 2007). Figure 3 shows the ANN neural architecture for the input-hidden-output interaction.
In Back Propagation method, the output of neurons in input layers serve as input to the hidden layer, while the output of the hidden layers serves as input for output layer that finally gives desired output .In this present study, the design matrix with the experimental generated responses (Table 4) were used to develop a robust network structure. Learning choice of 70-20-10% was adopted for network training, testing and validation.
The network node activation was done using sigmoid transfer function (STF) as shown in Eq. 1.
The network training was conducted using Levenberg-Marquardt learning algorithm (Trainlm) and the output function followed Purelin transfer function (Venkata et al. 2015). ANN-Genetic algorithm was used for process optimization.

Characterization of paint wastewater
The results obtained from the characterization of the paint wastewater sample are presented in Table 5. From Table 5, it can be observed that the paint wastewater has high content of TSS and TDS of 2685 mg/l and 1318 mg/l, respectively as against the NERS (National effluent regulatory standard) of 705 mg/l and 1200 mg/l (Ezemagu et al. 2021a;). The high TSS and TDS observed in paint wastewater indicate that the wastewater contains high particle load. Comparing the total solid content with NERS, it was observed that the paint wastewater sample contains 2098 mg/l in excess of the NERS. Hence it can be inferred that the paint wastewater is highly turbid and cannot be discharged to the environment without treatment. In addition, it was observed that the sample's pH is within the NERS acceptable standard.

The proximate characterization of GSF
Proximate analysis was carried out on the gastropod shell flour (GSF) to determine the proximate compositions such  as crude protein, oil content, ash content, bulk density and moisture content using standard methods cited in Table 2. The result of the proximate analysis shows that GSF contains high quantity of crude protein (42%). Based on the protein content observed, it could be inferred that GSF is an efficient precursor for the extraction of raw protein (conchiolin) that can be used for wastewater treatment. The oil content was found to be 7.4%. Oil content of this percentage (< 10%) would have negligible inhibitory effect on the deprotenization process (Ezemagu et al. 2021a). The total yield of 86% was obtained indicating 14% weight loss which could be attributed to the volatile components present in the GSF sample. The bulk density of 0.33 g/ml indicates that GSF is extremely aeratable. In addition, the ash content of 10% shows that the flour is rich in minerals while negligible moisture content of 8.6% was obtained.

Elemental analysis
Elemental analysis was carried out on both the GSF and the GSC to evaluate their qualitative and quantitative composition. The results obtained are reported in Table 6. From the results, it could be seen that GSF has high content of calcium. Calcium content of 70% was recorded which supports the claim that GSF contains between 70 and 98% calcium (Rao et al. 2007). This high calcium  content is justified since the animal uses it for body replenishment (Rao et al. 2007). High content of oxygen (26.36) in GSF can be traced to the presence of protein in the shell (Ezemagu et al. 2021a). The presence of carbon (7.47%) is attributed to the carbonaceous nature of GSF. From the elemental characterization of GSC (Table 6), it could be observed that more elements were present (Na, Mg, Al. P, Si, Cl, K) which were not in GSF. The additional elements and the observed 20.7% reduction in calcium content in GSC can be attributed to the effect of GSF reaction with the extraction solution. The oxygen content of both GSF and GSC were observed to be approximately the same (26.36 and 25.94%), indicating that the deproteinization process was effective.

FTIR studies
The infrared spectra of GSF and GSC shown in Figs. 4 and 5 revealed peaks representing different functional groups. It is observable that the spectra fall within the mid infrared region (4000-400 cm −1 ). Figures 4 and 5 are analyzed and compared with the existing FTIR data base (FDM NIST08 Mass Spectral Library) (Ezemagu et al. 2021a; Okey-Onyesolu et al. 2020; Rao et al. 2007). From the regions of absorbance, some functional groups were observed. The FTIR spectrum pattern for GSF (Fig. 4) exhibits 20 discernable peaks at frequency of 4000-700 cm −1 , threshold of 0.44; while in that of GSC, 16 discernable peaks were observed (Fig. 5) between the frequencies of 4000-600 cm −1 . The principal peaks in the spectrum were detected at 3648 cm −1 , 3627 cm −1 , 3580 cm −1 , 3291 cm −1 , 1457 cm −1 , 1082 cm −1 , 1017 cm −1 , 844 cm −1 , 712 cm −1 and 700 cm −1 (Fig. 4). The highest peak at 1457 cm −1 was observed within the FTIR fingerprint region. The presence of aromatic group was exhibited by the broad bands in the regions above 3000 cm −1 (3281 cm −1 , 3580 cm −1 , 3627 cm −1 and 3648 cm −1 ). The peak at 2919 cm −1 shows the presence of asymmetric methyl group, peaks at 1082 cm −1 and 1017 cm −1 depict the aliphatic C-N stretching while peaks at 844 cm −1 , 712 cm −1 and 700 cm −1 show the presence of phosphorous compound of P-F stretching. Figure 5 shows distinct peaks for GSC. The reduction in number of peaks when compared with Fig. 4 shows that some functional groups were removed during the extraction process. A shift in peaks orientation can also be observed from the X-H stretching region to fingerprint region. In  Fig. 4, 14 discernable peaks were found within the X-H stretching region, while 2 peaks were observed within the same region in Fig. 5. The shift can be associated with longitudinal acoustical modes (accordion modes) resulting from molecular distortion and bond breaking during the extraction (Rao et al. 2007;Venkata et al. 2015). The stunted broad band between 4000 and 3000 cm −1 in Fig. 4 is replaced with very broad strong band (in Fig. 5) at 3645 cm −1 and 3356 cm −1 indicating Si-OH stretching which can be confused with those of O-H frequencies. The highest peak on Fig. 5 is observed at 1456 cm −1 , with threshold frequency of 1.08. The sharp distinct peak at 1456 cm −1 can be connected to methylene scissoring in alkane group. Also, discernable peaks were recorded at the upper wave number end, the peaks at 699 cm −1 , and 648 cm −1 are linked to C-H bending of alkyne group. The peak observed at 1082 cm −1 is an indication of C-O stretching band (Ethers) due to the C-O-C linkage. C-O stretching band can be observed near 1150 cm −1 (1154 cm −1 ), indicating the presence of Anhydrides. A broad N-H wagging band also appears at 750-650 cm −1 indicating the presence of secondary amine which can be linked to the protein constituent of GSC.

X-Ray diffraction analysis of GSF and GSC
The X-ray diffraction spectrum of GSF and GSC are shown in Figs. 6 and 7. It can be observed that Fig. 6 clearly shows well recognized intense peaks. This spectrum is an X-Y plot of 2θ vs X-ray count (intensity). Fourteen clear peaks assigned due to their different reflections and planes were observed at scattering angles of 2θ = 26.5°, 27.8°, 31.5°, 33°, 36°, 37°, 38°, 42°, 43°, 46°, 48°, 51°, 52.5°, 53°. From the nature of these peaks in Fig. 6, a symmetric organized crystalline structure can be inferred. The spectrum for GSC presented in Fig. 7 shows a less coherent arrangement of fourteen distinct peaks when compared with Fig. 5. The peaks can be observed at scattering angles of 2θ = 31.5°, 34°, 35°, 36.5°, 38°, 42°, 43°, 45°, 46°, 48°, 51° (Fig. 7). The asymmetric peaks arrangement in Fig. 7 indicates a semicrystalline molecular arrangement. This type of molecular arrangement infers that GSC is an isotropic amorphous compound (Takagi et al. 2004). Comparison between Figs. 6 and 7 based on the nature of peaks show that GSF is more structurally stable than GSC.

DSC/TGA analysis for GSF and GSC
The DSC and TGA representation of GSF and GSC obtained are shown below in Fig. 8a-d, respectively. Figure 8a, b (the DSC profiles of GSF and GSC) represent application of DSC for the characterization of the phase transition that occurred in GSF and GSC over the temperature ranges of 38-298 °C and 45-300 °C, respectively. The transition enthalpies of 23.091 kJ/mol and 11.620 kJ/mol, respectively were obtained. The thermal activation energy (ΔE) was evaluated through TGA to be 25.86 kJ/mol and 45.928 kJ/mol for GSF and GSC, respectively using method described by Rao etal. (2007). GSF produced sharp transition in the temperature range of 62.5-81 °C, while GSC produced its sharp transition between 49 and 52 °C. These behaviors could be linked to spontaneous densification during thermal treatment of the samples. The densification of the aggregated mass took place at temperatures of 100-150 °C for GSF (Fig. 8a) and 115-175 °C for GSC (Fig. 8b). The glass transition temperatures were observed between 37.5 and 42 °C for GSF and 48-52 °C for GSC. Furthermore, it was observed that GSC has higher glass transition temperature than GSF; this implies that GSC can withstand more operational increase in temperature than GSF without being denatured (Rao et al. 2007;Clark and Pitt 2008). Within the glass transition stage, the onset, midpoint and offset transition points can be observed at the temperatures of 37.5 °C, 39.3 °C and 42 °C for GSF, and 48 °C, 50 °C and 52 °C for GSC, respectively. A clear observation of the DSC graphs demonstrated a situation in which the heat flow discs indicate exothermic nature for both GSF and GSC. Figure 8c, d shows the thermal-gravimetric analysis (TGA) profiles of the two samples (GSF and GSC). Graphically, the Figures represent variation in weight with respect to temperature. The final residual masses for GSF and GSC estimated based on weight loss with respect to temperature are 5.72218 mg and 1.974 mg, representing 89.2% and 74.9% of the original weights of GSF (6.415 mg) and GSC (2.634 mg) sample, respectively. The initial weight loss observed in Fig. 8c, d could be linked to internal moisture content and gaseous loss from the matrix molecules (Rao et al. 2007). The second phase weight loss may be as a result of decomposition in the samples. The results conclusively suggested thermal operational stability of the GSF and GSC as indicated by significant final residue percentage of 89.2% and 74.9% for GSF and GSC, respectively.

SEM characterization of GSC
The analysis of the external morphology (texture) of GSC was obtained via scanning electron microscopic evaluation (SEM). The SEM image obtained is presented in Fig. 9. A compact structure with tiny pores and small stick littered external morphology is observable at 100 μm. It shows a good characteristic for an effective coagulant. Coagulants with reduced particle sizes and increased surface porosity would provide a better platform for adsorption of fine suspended particles. This result is similar to the report of Bahrodin et al. (2021).

Effect of process variables via OFAT
Conventional application of one factor at a time (OFAT) method evaluates the impact of one variable within a process by holding other variables at constant level (Khusro 2016).

Effect of pH on coagulation efficiency
The effect of pH on the coagulation system under consideration is illustrated in Fig. 10. The pH of the solution is a critical parameter in most treatment processes Venkata et al. 2015). It affects the surface charge of the coagulant (Zhao et al. 2011). From Fig. 10, at constant GSC dosage of 5 g and 30 min settling time, two regions are notable, a region where the removal efficiency was at maximum and a region where it was at minimum. The removal efficiency was highest at pH of 4. A similar result was reported by Zhao et al. (2015) on the effect of pH on coagulation, using ferric-based coagulant in yellow river water treatment. Also, Sun et al. (2019) and Vuppala et al. (2019) reported best removal efficiency within the same acidic region. The high removal efficiency within 2-4 could be attributed to progressive protonation of the coagulation system as GSC releases positive charges which progressively conjugated with the available negative species toward equilibrium. The region between the pH of 5-8 represents a region of progressive decline in particle neutralization and floc formation. This decline can be attributed to decline in GSC solubility within the system as a result of change in pH from strong acidity to alkalinity (Menkiti et al. 2018). At pH of 8, minimum particle removal efficiency ( < 12%) was observed. This pH of minimum particle removal may be referred to as the point of zero charge for GSC. At this point (pH of 8), the surface charges of the coagulant available for charge neutralization are negligible. After the point of zero charge of GSC, the coagulation environment becomes more alkaline. At this stage, GSC is less soluble. However, the surface charge is not completely zero, which  (Zhao et al. 2015). This result ("Effect of pH on coagulation efficiency" section) suggests that GSC is more effective in acidic environment.

Effect of coagulant dosage on coagulation efficiency
Variation in turbidity removal efficiency with GSC dosage at constant time (30 min) and best pH (4) adopted from "FTIR studies" section is presented in Fig. 11. Turbidity removal efficiency was found to increase from minimum to the maximum with increase in GSC dosage before a sharp decline (Fig. 11) similar to the reports of Ezemagu et al. (2021a) and Bahrodin et al. (2021)). This trend (progressive increase of removal efficiency to maximum) could be as a result of increase on the availability of positively charged particles provided by the GSC for destabilization of the negatively charged suspended particles. A sharp decline in particles removal efficiency from 4 g/l could be attributed to excess coagulant concentration, which may have resulted to excess protonation (increased net concentration of positively charged coagulant particles). Menkiti et al. (2018) opted that such protonation will cause re-turbidization of the effluent. The curved profile of Fig. 11 shows the effect of individual coagulant dosage on the particles decontamination process. The minimum particles decontamination efficiency (18%) was observed at the coagulant dosage of 0.5 g/l, while the highest particles removal efficiency (98.1%) was observed at 4 g/l of GSC.

Effect of settling time on coagulation efficiency at different temperatures
The variations in removal efficiency with time at different temperatures are shown in Fig. 12. It was observed that removal efficiency increased with settling time till equilibrium was attained. This is similar to the report of Chi and Cheng (2006). At equilibrium stage, 93.1%, 95.3%, and 98.7% reduction in turbidity was achieved at 25 °C, 35 °C and 45 °C, respectively. The equilibrium stages were observed at 20 min after which there was no significant reduction in turbidity of the treated paint wastewater samples. Hence, most of the flocs settled between 0 and 20 min. Menkiti et al. (2018) suggested that equilibrium time between 15 and 20 min is desirable for peri-kinetic coagulation. In addition, it was also observed that coagulation efficiency increased with temperature, while the samples viscosity decreased. This trend can be attributed to molecular excitation due to increase in particles kinetic energy (Rao et al. 2007). Zhang et al. (2022) reported that increase in particles kinetic energy enhances the interactive motion of the particles creating more effective particles collision and floc formation. Highest efficiency of 93.16% was observed at 25 °C. Slight increase was observed (from 93.16 to 95.433%) as the temperature increased to 35 °C. Furthermore, the removal efficiency increased to 98.724% as the temperature was increased to 45 °C. Many researchers also reported high removal efficiencies within the same temperature range (Liu and Zhou 2004;Ejimofor et al. 2020bEjimofor et al. , 2021bCheng et al. 2005;Nwabanne et al. 2018).

Characteristics of the treated water after coagulation
After the coagulation experiment, the treat wastewater had TS of 791 ± 0.03 mg/l, TSS of 275 ± 0.024, TDS of 516 ± 0.016 and pH of 7.2 as against the national discharge limit of 1905 mg/l, 705 mg/l, 1200 and pH 7-8. Hence, it can be inferred that the treated wastewater can be discharged into the environment without adverse effect. In addition, it implies that the use of GSC achieved TS, TSS and TDS removal of about. Figure 13 shows the artificial neural network system histogram. ANN system histogram represents the error that approaches the network mean square error (MSE). It shows that approximately 12 instances were used for training, 5 for testing while the remaining was used for validation. The zero error line reveals -6.202 errors in 20 bins. The error shows the extent of correlation between the target and prediction. Hence, the negative sign show that predictions are higher than target. However, error of values < 10 indicates relatively good correlation between the target and the predicted responses. The network progress was monitored based on validation error.

Response prediction and optimization using ANN and GA
Network training was terminated at increase in validation error. The root-mean-squared error (RMSE) network performance curve shown in Fig. 14 illustrates the plot of the observed RMSE against the epochs. In Fig. 14, three different curves were built for training, testing and validation. The dotted line shows the best possible network conditions for training, testing and network validation based on the root-mean-squared error. The best validation performance was obtained at RMSE of 7.7084 at epoch 5. The validation performance suggests that after the fifth iterations (epoch 5), the network attained its best learning stage at the lowest possible RMSE (7.7084). Also, the RMSE measures the correlation between the target and ANN predicted responses. The best RMSE < 10 infers relative high correlation between the output (ANN predicted turbidity removal) and the target (Experimental turbidity removal). In addition, the network capacity to predict significant output was illustrated by the training state plot (Fig. 15).
The network gradient of 1.9506 −11 (loss function) was computed to illustrate the error contribution of each neuron at 5 epochs (Fig. 15). Lower error is better (Ezemagu et al. 2021a). The gradient was very much less than unity (1.9506e−11), it indicates that the error contribution of each neuron within the 5 epochs is minimal. Momentum gain (Mu) is the training gains and its value is expected to be less than one (Ezemagu et al. 2021a) "Mu" closer to zero shows high capacity of the trained network in making significant response prediction. In support of the RMSE and the training output result in network accuracy, the training, testing, validation and the overall network correlation coefficient (R 2 ) of 0.97, 0.99, 0.97 and 0.93 were obtained from the performance plot of training, testing, validation and All (overall network performance) as shown in Fig. 16.
Based on the R 2 -values, (0.97, 0.99, 0.97 and 0.93), it is inferred that the network has significant fitting performances in all the stages (training, testing and validation). The overall network performance equation (Eq. 2) shows the overall relationship between the target and the output. The network output (predicted Turbidity removal efficiency (%)) obtained from software iteration within the hidden layers (Table not shown) was compared with the target (experimental removal efficiency) based on the correlation coefficient. A strong correlation is always inferred for R 2 value ≥ 0.7 (Rao et al. 2007).Correlation coefficient of 0.92 and mean average percentage error (MAV.PE) of 0.36 were obtained (Low average percentage error (close to zero) is desired because it indicates insignificant error between the experimental and predicted data). From the present comparison very strong correlation between the two data sets (output and target) can be inferred. In support of this judgment based on R 2 values and the MAV.PE, Fig. 17 shows the plot of (2) Output = 0.73 * target + 24 output and target with respect to experimental run order. Visual assessment of Fig. 17 gave credence to the result obtained in Fig. 14. The slight deviations observed at run 14 and 20 (Fig. 17) may be attributed to data approximation during generation of hidden hyper-parameters for response prediction.
The optimization of turbidity removal efficiency based on ANN predicted overall model equation (Eq. 2) was carried out through genetic algorithm (GA). This metaheuristic AI approach provides an approximate optimal by evaluating sets of chromosomes created via GA operational environment. The target of the response optimization (set at 99.999%) is subject to the total weight such that the optimal response is as higher as possible and not exceeding the maximum weight (Ezemagu et al. 2021a;Rao et al. 2007;Vilardi 2020) which is 100% for the present study. The weight of the data set was scaled into a continuous interval using fuzzy C-means. Population data in Table 4 applied as the initial population to Eq. 2 (transfer function relating the weight to target) passed through reoccurring data crossover and mutation before termination. The outcome of the GA operation based on the convergence of optimal solutions show that the optimal turbidity removal of 98% can be accomplished at GSC dosage, time and temperature of 4 g, 20 min and 45 °C, respectively. This optimal response was selected from a queue of 20 local optima conditions (not shown) returned at termination of GA operation.

Conclusion
In this study, gastropod shell conchiolin which was a waste product was successfully processed and used as natural coagulant for particle decontamination and turbidity removal from paint wastewater. Conchiolin showed remarkable turbidity removal efficiency at slightly acidic pH. Also, coagulant dosages, settling time and temperature influenced the turbidity removal. Hybridization of response surface methodology and artificial neural network-genetic algorithm for design, prediction and optimization of the wastewater treatment conditions were successfully carried out. The results show that response surface design predictions can be successfully optimized using artificial neural network-genetic algorithm. An optimized turbidity removal efficiency of 98.6% was obtained at 4 g gastropod shell conchiolin, 20 min and 45 °C. Practicably, these conditions are desirable for peri-kinetic coagulation treatment and can be adopted for design of wastewater treatment plants.