Paper · MFC
Electricity generation from renewable source- A leachate based microbial fuel cell machine learning approach
Summary
What the paper reports, in its own words.
Abstract
Abstract This research presents an innovative blend of feature selection and sensitivity analysis techniques, which is an essential yet overlooked aspect in the study of MFCs. The study compared predictive models utilizing various machine learning algorithms to assess the impact of time, dosage, pH and temperature on ammonium nitrogen concentration (NH 4 -N) to predict the power density (PD) output of microbial fuel cells using leachate as a substrate for treatment. Evaluation of six machine learning models demonstrates varying levels of predictive accuracy. CatBoost (R2:0.9969, MSE: 48.8430, RMSE:6.9888) emerges as the most accurate model, followed closely by XGBoost (R2:0.9917, MSE:130.1668, RMSE:11.4091) and Random Forest (R2:0.9830, MSE:267.0929, RMSE:16.3430). Time series plots illustrate the performance of different models in predicting PD over a period, indicating good alignment with observed data. Comparison of Mean Squared Error (MSE) highlights significant variations in prediction accuracy, with CatBoost demonstrating the greatest enhancement and precision. The study directly tackles the deficiencies in existing MFC predictive modeling by incorporating the CatBoost algorithm, which provides enhanced accuracy and a deeper understanding of the nonlinear connections between environmental variables and power density.
AI summary
55% confidenceThis paper applies machine learning (CatBoost, XGBoost, Random Forest) to predict power density in H-type dual-chamber MFCs treating landfill leachate, using time, pH, temperature, and NH₄-N concentration as inputs. CatBoost achieved R² = 0.9969 with RMSE = 6.99 mW/m², outperforming other ensemble methods. The study emphasizes feature selection and sensitivity analysis but provides no mechanistic insight into nitrogen cycling, electrode potentials, current density operating windows, or N₂O/NO₂⁻ accumulation—critical parameters for bioelectrochemical nitrogen management.
Key insights
Mechanism NH₄-N and temperature show moderate positive correlation (r = 0.58), suggesting temperature variations directly impact ammonia dynamics in MFC environments.
So what. Operators should monitor and control temperature as a lever to manage ammonia concentration; temperature swings may destabilize nitrogen speciation.
Performance CatBoost (R² 0.9969, RMSE 6.99) outperforms XGBoost (R² 0.9917, RMSE 11.41) and Random Forest (R² 0.9830, RMSE 16.34) by 0.52% and 1.39% respectively in R² on this dataset.
So what. For leachate-fed MFCs with this input feature set, CatBoost is the preferred predictive model; however, the marginal gain (1.4%) may not justify added complexity if RF is already deployed.
Performance Optimal pH range for power generation is 6.5–7.5 (up to 1200 mW/m²), while hydrogen generation peaks at pH 5.0–6.0 (8 m³ H₂/m³/d); dataset pH range is 6.85–7.0, entirely within the power-optimized window.
So what. The narrow pH window (6.85–7.0) in this study avoids the hydrogen-favoring regime; operators should test pH <6.5 if ammonia recovery or hydrogen co-production is desired.
Threshold Optimal microbial activity occurs at 35–40°C; dataset mean temperature is 30.21°C (range 27.31–33.42°C), below the optimum, yet power density still reaches 339.41 mW/m².
So what. Heating to 35–40°C could boost power output; current sub-optimal temperature may be masking the system's true potential or may reflect ambient conditions in the study location.
Opportunity areas6
Specific gaps in this paper, with an experiment that would close each.
High No measurement or prediction of electrode potentials (anode, cathode, cell voltage) or current density; these are essential for diagnosing nitrogen redox pathways and nitrification vs. denitrification dominance.
Paper measures only power density and NH₄-N; electrode potential controls which nitrogen species (NO₃⁻, NO₂⁻, N₂O, N₂) are produced or consumed. Current density determines mass transfer and biofilm kinetics.
Try. Measure anode and cathode potentials (vs. Ag/AgCl) and current density (A/m²) across the range of NH₄-N (0–300 mg/L) and temperature (27–40°C); correlate with outlet nitrogen speciation.
High No quantification of nitrogen speciation in effluent (NO₃⁻, NO₂⁻, N₂O, N₂, residual NH₄⁺); ammonia recovery or denitrification efficiency not assessed.
Paper tracks only inlet NH₄-N; no outlet analysis. Leachate treatment requires knowing whether nitrogen is oxidized (nitrification), reduced (denitrification), or recovered (stripping/precipitation).
Try. Measure outlet NO₃⁻-N, NO₂⁻-N, N₂O (headspace gas chromatography), and residual NH₄⁺-N at steady state for each condition; calculate nitrogen mass balance and recovery/removal efficiency.
High No exploration of C/N ratio or organic loading rate (COD/NH₄-N) effects on power density and nitrogen cycling; only COD removal mentioned in prior work (Ishaq et al., 2024).
Paper states 'Nitrogen can alter nutrients for microbial growth and pollutant removal' but does not vary C/N ratio. Leachate C/N is typically low (1–3), which may limit denitrification or favor nitrification.
Try. Vary C/N ratio from 1 to 10 by adjusting acetate or glucose co-feed; measure power density, NH₄-N removal, and outlet NO₃⁻/NO₂⁻ to identify optimal C/N for energy recovery vs. nitrogen removal.
High No investigation of dissolved oxygen (DO) in cathode compartment or its effect on nitrification vs. denitrification; aerated cathode is mentioned but DO not controlled or measured.
Paper states 'aerated cathode compartments' but provides no DO setpoint, measurement, or sensitivity analysis. DO is the primary driver of nitrification kinetics and N₂O production in bioelectrochemical systems.
Try. Vary cathode DO (0–100% saturation) while monitoring anode potential, power density, and outlet NO₃⁻-N and N₂O; identify DO threshold for nitrification onset and N₂O accumulation.
Medium No analysis of HRT (hydraulic retention time) or batch vs. continuous operation; time is an input feature but only as absolute seconds, not normalized to HRT.
Table 1 shows time range 60–216,000 s (~1 min to 60 h), but no HRT definition or discussion. Nitrogen cycling kinetics depend on HRT; short HRT favors power output, long HRT favors complete nitrification/denitrification.
Try. Operate at fixed HRT (e.g., 12, 24, 48 h) across the temperature and NH₄-N ranges; measure power density and nitrogen speciation to quantify HRT trade-off.
Medium No measurement of catholyte pH or its effect on nitrification kinetics; only anode pH (6.85–7.0) is reported, and no separate cathode pH control.
Paper notes 'pH directly influences electron and proton generation' but does not measure or vary cathode pH. Nitrification is pH-sensitive (optimum ~7.5–8.0); cathode pH may drift due to proton consumption.
Try. Measure and control cathode pH independently (e.g., 6.5, 7.0, 7.5, 8.0); quantify effect on nitrification rate, N₂O production, and power density.
Generated by MESSAI extraction pipeline · review against source PDF
Extracted values
Every value pulled from the paper, with its source, so each number can be checked against the text.
n = 47 extracted values
Select a citation to see the quote, page and confidence behind a value.
17 of 47 extracted rows modelable
- 18 not canonicalised (no canonical slug)
- 11 no SI value
- 1 contradicts other papers
| Parameter | Value | Unit | Confidence | Source | Notes | Modelable |
|---|---|---|---|---|---|---|
| Electrode Surface Area | 1.71 | cm | 95% | extracted | "Regarding the electrodes, carbon felt with the following dimensions: height (h) …" · condition_hint=design parameter for electrode dimensions, electrode_material=carbon felt | Excluded: no SI value |
| Validation dataset fraction | 10 | % | 95% | extracted | "The dataset consisted of about 1000 samples, which were divided into three separ…" · condition_hint=machine learning model validation configuration | Excluded: not canonicalised (no canonical slug) |
| Testing dataset fraction | 20 | % | 95% | extracted | "The dataset consisted of about 1000 samples, which were divided into three separ…" · condition_hint=machine learning model testing configuration | Excluded: not canonicalised (no canonical slug) |
| Internal Resistance | 1,000 | Ω | 95% | extracted | "Therefore, 1 kilo-ohm resistors were employed in conjunction with a lab jack U3-…" · substrate=leachate | Modelable |
| Training Data Size | 70 | % | 95% | extracted | "The dataset consisted of about 1000 samples, which were divided into three separ…" · condition_hint=machine learning model training configuration | Excluded: no SI value |
| Reactor Volume | 100 | ml | 95% | extracted | "Each microbial fuel cell (MFC) had a cathode chamber and an anode chamber, with …" · condition_hint=design parameter for MFC reactor volume, electrode_material=carbon felt | Modelable |
| Electrode Thickness | 1.71 | cm | 95% | extracted | "Regarding the electrodes, carbon felt with the following dimensions: height (h) …" · condition_hint=design parameter for electrode dimensions, electrode_material=carbon felt | Modelable |
| Electrode Thickness | 0.62 | cm | 95% | extracted | "Regarding the electrodes, carbon felt with the following dimensions: height (h) …" · condition_hint=design parameter for electrode dimensions, electrode_material=carbon felt | Modelable |
| Nafion membrane pretreatment temperature | 60 | °C | 95% | extracted | "The Nafion 117 membrane, manufactured by DuPont in Delaware, USA, underwent pre-…" · T=60°C | Excluded: not canonicalised (no canonical slug) |
| Electrode acetone immersion duration | 24 | h | 95% | extracted | "The anode and cathode were completely submerged in acetone for a duration of 24 …" · condition_hint=design parameter for electrode pretreatment, electrode_material=carbon felt | Excluded: not canonicalised (no canonical slug) |
| Nafion membrane pretreatment duration | 60 | min | 95% | extracted | "The Nafion 117 membrane, manufactured by DuPont in Delaware, USA, underwent pre-…" · T=60°C | Excluded: not canonicalised (no canonical slug) |
| R-squared (Random Forest) | 0.983 | — | 90% | random forest machine learning algorithm | "Table 3 Performance analysis for PD based on implication Models R2 MSE RMSE RF 0…" · substrate=leachate | Excluded: no SI value |
| Voltage Monitoring Interval | 60 | s | 90% | voltage measurement at regular intervals | "The voltage potential was measured at regular 60-second intervals without the pr…" · substrate=leachate | Modelable |
| Mean Squared Error (AdaBoost) | 1,880 | — | 90% | adaboost machine learning algorithm | "Table 3 Performance analysis for PD based on implication Models R2 MSE RMSE RF 0…" · substrate=leachate | Excluded: not canonicalised (no canonical slug) |
| Root Mean Squared Error (XGBoost) | 11.4 | — | 90% | xgboost machine learning algorithm | "Table 3 Performance analysis for PD based on implication Models R2 MSE RMSE RF 0…" · substrate=leachate | Excluded: not canonicalised (no canonical slug) |
| Root Mean Squared Error (Decision Tree) | 39.5 | — | 90% | decision tree machine learning algorithm | "Table 3 Performance analysis for PD based on implication Models R2 MSE RMSE RF 0…" · substrate=leachate | Excluded: not canonicalised (no canonical slug) |
| Mean Squared Error (CatBoost) | 48.8 | — | 90% | catboost machine learning algorithm | "Table 3 Performance analysis for PD based on implication Models R2 MSE RMSE RF 0…" · substrate=leachate | Excluded: not canonicalised (no canonical slug) |
| Root Mean Squared Error (Gradient Boosting) | 18.7 | — | 90% | gradient boosting machine learning algorithm | "Table 3 Performance analysis for PD based on implication Models R2 MSE RMSE RF 0…" · substrate=leachate | Excluded: not canonicalised (no canonical slug) |
| Electrode Surface Area | 4 | cm | 90% | extracted | "The electrodes were positioned at a distance of roughly 4cm from each other, wit…" · condition_hint=design parameter for electrode separation, electrode_material=carbon felt | Excluded: no SI value |
| R-squared (Decision Tree) | 0.901 | — | 90% | decision tree machine learning algorithm | "Table 3 Performance analysis for PD based on implication Models R2 MSE RMSE RF 0…" · substrate=leachate | Excluded: no SI value |
| Mean Squared Error (Decision Tree) | 1,560 | — | 90% | decision tree machine learning algorithm | "Table 3 Performance analysis for PD based on implication Models R2 MSE RMSE RF 0…" · substrate=leachate | Excluded: not canonicalised (no canonical slug) |
| R-squared (Gradient Boosting) | 0.978 | — | 90% | gradient boosting machine learning algorithm | "Table 3 Performance analysis for PD based on implication Models R2 MSE RMSE RF 0…" · substrate=leachate | Excluded: no SI value |
| R-squared (AdaBoost) | 0.88 | — | 90% | adaboost machine learning algorithm | "Table 3 Performance analysis for PD based on implication Models R2 MSE RMSE RF 0…" · substrate=leachate | Excluded: no SI value |
| Root Mean Squared Error (AdaBoost) | 43.4 | — | 90% | adaboost machine learning algorithm | "Table 3 Performance analysis for PD based on implication Models R2 MSE RMSE RF 0…" · substrate=leachate | Excluded: not canonicalised (no canonical slug) |
| Root Mean Squared Error (Random Forest) | 16.3 | — | 90% | random forest machine learning algorithm | "Table 3 Performance analysis for PD based on implication Models R2 MSE RMSE RF 0…" · substrate=leachate | Excluded: not canonicalised (no canonical slug) |
Methodology
What was the system and how was it run
Methodology
Methodology
The study used H-type dual-chamber MFCs (100 mL total, carbon felt electrodes, Nafion 117 PEM) fed with landfill leachate. A dataset of ~1000 samples (70% training, 10% validation, 20% test) was generated by varying time, pH (6.85–7.0), temperature (27.31–33.42°C), and NH₄-N (0.00159–283.38 mg/L), with power density as the output. Six ensemble machine learning models (Random Forest, Gradient Boosting, AdaBoost, Decision Tree, XGBoost, CatBoost) were trained and evaluated using R², MSE, and RMSE metrics. Feature selection and sensitivity analysis (correlation heatmap) were performed to identify key drivers.
System configuration
| Field | Value |
|---|---|
| system type | MFC |
Operating conditions
17 runs extracted| Run | T°C | pHan | pHcat | Substrate | [S]mg/L | HRTh | Mode | R_extΩ | V_appV | Inoculum |
|---|---|---|---|---|---|---|---|---|---|---|
| v2-0ed04b56fd65fcc6 | — | — | — | leachate | — | — | — | — | — | — |
| v2-26cf2f8771fe2ddc | 40 | — | — | — | — | — | — | — | — | — |
| v2-39bf2345ce18e6a3 | — | 7 | — | leachate | — | — | — | — | — | — |
| v2-44131bc9d3dbc039 | — | — | — | leachate | — | — | — | 1000 | — | — |
| v2-51774903b65f74f4 | — | 6.88 | — | leachate | — | — | — | — | — | — |
| v2-547db789816d2309 | 33.42 | — | — | leachate | — | — | — | — | — | — |
| v2-5b5e39b30f9c5574 | 27.31 | — | — | — | — | — | — | — | — | — |
| v2-73d6ff2b61e30505 | 30.21 | 6.88 | — | leachate | 84.43 | — | steady_state | 1000 | — | — |
| v2-7606e625f8f4e35c | 30.21 | 6.85 | — | leachate | 84.43 | — | — | — | — | — |
| v2-8ba51b05ccbc6251 | 30.21 | — | — | leachate | — | — | — | — | — | — |
| v2-a1dae8e5b0f32bc2 | 60 | — | — | — | — | — | — | — | — | — |
| v2-b059e5647a30a50a | 33.42 | — | — | — | — | — | — | — | — | — |
| v2-ce7e049b84175adf | 30.21 | 6.88 | — | leachate | 84.43 | — | — | — | — | — |
| v2-d70574efcbfd766b | — | 6.85 | — | leachate | — | — | — | — | — | — |
| v2-deb99d525b1bbb37 | 27.31 | — | — | leachate | — | — | — | — | — | — |
| v2-df36c50fcf80618e | 30.21 | — | — | — | — | — | — | — | — | — |
| v2-f586e9af22985397 | — | — | — | — | — | — | — | — | — | — |
Pipeline
Raw extractions, artifacts, quality signal
Extraction quality
honesty signalCalibrate your trust in the extracted content against these signals — they're how the pipeline rated its own output for this paper.
Parameter completeness
13%
Fraction of expected parameters extracted
Data completeness
40%
Overall extraction coverage across fields
AI confidence
55%
Model self-reported confidence
Mean parameter confidence
89%
Average confidence across PaperParameter rows surfaced above
review against source PDF before citing · scores 0–1
Chart explorer
Move the inputs this paper used and see how the predicted outcome shifts.
Explore model
The reactor in 3D, with flow you can switch on and off.
AI insights
Contradictions, gaps and opportunities found by comparing this paper with the corpus.
Compare similar papers
The nearest papers by system, materials and organisms, with their outcomes side by side.
Source PDF
The paper as published.
Representative model, generic MFC
Tap to interactRepresentative model, generic MFC
Representative MFC — matched on the paper’s system type only, not its reactor or geometry.
Keywords
- Microbial fuel cell
- Renewable energy
- Electricity
- Electricity generation
- Fuel cells
- Leachate
Cite
Shamsuddeen Jumande Mohammad, Aliyu Ishaq (2024). Electricity generation from renewable source- A leachate based microbial fuel cell machine learning approach. Research Square. https://doi.org/10.21203/rs.3.rs-5119060/v1
- Journal
- Research Square
Open in lab for full controls, parameter editing, and template overlays.
Open in lab