Machine Learning Framework for Urban Water Demand Forecasting: A Wavelet-Enhanced Support Vector Regression Ensemble

<p>Accurate water level (WL) forecasting is important for water resources management and planning purposes in the Great Lakes. The objectives of this research are two-fold.  The first objective is to apply machine learning (ML) (i.e., random forest (RF) and support vector regression (SVR)) and hybrid convolutional neural network(CNN)-long-short term memory (LSTM) deep learning (DL) models for multi-step (i.e., one-, two- and three-monthly step ahead) WL forecasting in the Great Lakes (Michigan and Ontario). The second objective is to integrate the boundary corrected (BC) maximal overlap discrete wavelet transform (MODWT) with SVR, RF, and CNN-LSTM models to improve the performance of the individual models. By employing a BC-wavelet decomposition method, the ‘future data’ issue (i.e., data from the future that is not available), often overlooked in the literature and a major barrier to achieving realistic forecasting performance is overcome. </p><p>For Lakes Michigan and Ontario, 1212 monthly WL (m) records (spanning Jan 1918–Dec 2018) were used to develop the models. For the non-wavelet-based models (SVR, RF, and CNN-LSTM), candidate model inputs included the WL recorded over the previous 12 months.  For the BC-MODWT-based models (BC-MODWT-SVR, BC-MODWT-RF, and BC-MODWT-CNN-LSTM), the lagged input time series were decomposed into BC-wavelet and scaling coefficients by using different mother wavelets (Haar, Daubechies, Symlets, Fejer-Korovkin and Coiflets), filter lengths (from two up to 12) and decomposition levels (from one up to seven).  For each method (SVR, RF, and CNN-LSTM), mother wavelet, and decomposition level a model was generated.  For both wavelet- and non-wavelet-based models, the particle swarm optimization (PSO) method was used to select the most appropriate inputs to include in the proposed multi-step WL forecasting models.</p><p>The datasets were partitioned into calibration and validation subsets. After calibrating the models, various performance evaluation metrics, e.g., coefficient of determination (R<sup>2</sup>), root mean square error (RMSE), mean absolute error (MAE), root mean square percentage error (RMSPE), mean absolute percentage error (MAPE) and the Nash-Sutcliffe efficiency coefficient (NSC) were used to assess model accuracy.</p><p>Of the ML models, the SVR outperformed RF while the DL models outperformed the ML models for each forecast lead time (one-, two-, and three-step(s) ahead). Results from this case study indicate that not all wavelet families and decomposition levels perform equally and in some cases, the wavelet-based models do not improve performance over the non-wavelet-based models. However, the BC-MODWT-CNN-LSTM using suitable mother wavelets (e.g., Haar) outperforms the individual ML and BC-MODWT-ML-based models. More accurate forecasts were obtained for Lake Michigan although the performance in both Great Lakes was accurate. The outcomes of this research indicate that the BC-MODWT-CNN-LSTM model is a promising tool for generating accurate WL forecasts.</p>

Accurate forecasts of water demand are a crucial factor in the strategic planning and judicious use of finite water resources within a region, underpinning sustainable socio-economic development. This study aims to compare the applicability of various artificial intelligence models for long-term water demand forecasting across different water use sectors. We utilized the Tuojiang River basin in Sichuan Province as our case study, comparing the performance of five artificial intelligence models: Genetic Algorithm optimized Back Propagation Neural Network (GA-BP), Extreme Learning Machine (ELM), Gaussian Process Regression (GPR), Support Vector Regression (SVR), and Random Forest (RF). These models were employed to predict water demand in the agricultural, industrial, domestic, and ecological sectors using actual water demand data and relevant influential factors from 2005 to 2020. Model performance was evaluated based on the Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE), with the most effective model used for 2025 water demand projections for each sector within the study area. Our findings reveal that the GPR model demonstrated superior results in predicting water demand for the agricultural, domestic, and ecological sectors, attaining R2 values of 0.9811, 0.9338, and 0.9142 for the respective test sets. Also, the GA-BP model performed optimally in predicting industrial water demand, with an R2 of 0.8580. The identified optimal prediction model provides a useful tool for future long-term water demand forecasting, promoting sustainable water resource management.

Background Leaf area is a crucial indicator of plant growth and physiology, with direct measurements being destructive to the plant. This study aimed to develop and compare machine learning models [support vector regression (SVR), adaptive neuro-fuzzy inference system (ANFIS), and deep multilayer perceptron (DMLP)] and linear regression (LRM) for the nondestructive prediction of leaf area in five colored cotton cultivars. A total of 1 334 leaves were sampled, and their length (L), width (W), and leaf area (LA) were determined via digitized images. The models were developed using 70% of the data for training and 30% for validation. Their performance was evaluated using the coefficient of determination ( R 2 ), root mean square error, mean absolute error, mean absolute percentage error, and Willmott's index of agreement. Results The results showed that the machine learning models, notably the ANFIS (triangular membership function), the DMLP (2–16-16–1 configuration), and the SVR [radial basis function (RBF) kernel], significantly outperformed the linear regression models in leaf area estimation accuracy. The ANFIS and DMLP models achieved the highest R 2 (0.979 3, test), followed by the SVR model ( R 2 = 0.979 0, test), all with minimal errors. Among the linear models, the LRM (using the L × W product) was the most effective ( R 2 = 0.978 3). Conclusions On the basis of the performance criteria of the models, the machine learning models are more accurate for the nondestructive estimation of leaf area in colored cotton. The best-performing model (SVR with RBF kernel) was made available in an interactive web application, aiming to optimize crop management with accurate and nondestructive data.

Tomato is one of the most important vegetables due to its high production and nutritional value. With the development of digital agriculture, the tomato breeding and processing industries have seen a rapid increase in the need for simple, low-labor, and inexpensive methods for analyzing tomato composition. This study proposes a digital method to predict four minerals (calcium, potassium, phosphorus, and magnesium) in beef-type tomato using machine learning models, including k-nearest neighbors (kNN), artificial neural networks (ANNs), and Support Vector Regression (SVR). The models were discriminated using the coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). The kNN model showed the best performance for estimation of quantity of calcium, potassium, phosphorus, and magnesium. The results demonstrate that kNN consistently outperforms ANNs and SVR across all target nutrients, achieving the highest R2 and the lowest error metrics (RMSE, MAE, and MAPE). Notably, kNN achieved an exceptional R2 of 0.8723 and a remarkably low MAPE of 3.95% in predicting phosphorus. This study highlights how machine learning can provide a versatile, accurate, and efficient solution for tomato mineral analysis in digital agriculture.

Hybrid random forest-based models for predicting shear strength of structural surfaces based on surface morphology parameters and metaheuristic algorithms

Accurate prediction of crop yield before harvest is critical to food security and importation. The calculated ten explanatory factors and autumn crop yield data were used as data sources in this research. Firstly, a Redundancy Analysis (RDA) was employed to carry out explanatory factors and feature selection. The simple effects of RDA were used to evaluate the interpretation rates of the explanatory factors. The conditional effects of RDA were adopted to select the features of the explanatory factors. Then, the autumn crop yield was divided into the training set and testing set with an 80/20 ratio, using Support Vector Regression (SVR), Random Forest Regression (RFR), and deep neural network (DNN) for the model, respectively. Finally, the coefficient of determination (R2), the root mean square error (RMSE), the mean absolute error (MAE), and the mean absolute percentage error (MAPE) were used to evaluate the performance of the model comprehensively. The results showed that the interpretation rates of the explanatory factors ranged from 54.3% to 85.0% (p = 0.002), which could reflect the autumn crop yields well. When a small number of sample training data (e.g., 80 samples) was used, the DNN model performed better than both SVR and RF models.

Owing to the significance of urban water resources demand management, and in accordance with urban water demand with features of both growth and periodicity, an urban water resources supply and demand management dynamic optimal control model was constructed. On the basis of the prediction of urban water demand, the urban water demand could be satisfied through reasonable control of urban water supply investment. Under state space constraint, a complementary slackness stronger form, required by the method of applying maximum principle to solving optimal control problem, was analyzed not always to be true on the conjoint point where optimal trajectory changes from active into inactive. Then, the normal form of complementary slackness condition should be applied to solving the problem under the circumstances. By implemented singular optimal control on the system, the total square of deviation of urban water resources supply and demand could be minimized in the period of urban water demand growth.

Recently, the quality of fresh water resources is threatened by numerous pollutants. Prediction of water quality is an important tool for controlling and reducing water pollution. By employing superior big data processing ability of deep learning it is possible to improve the accuracy of prediction. This paper proposes a method for predicting water quality based on the deep belief network (DBN) model. First, the particle swarm optimization (PSO) algorithm is used to optimize the network parameters of the deep belief network, which is to extract feature vectors of water quality time series data at multiple scales. Then, combined with the least squares support vector regression (LSSVR) machine which is taken as the top prediction layer of the model, a new water quality prediction model referred to as PSO-DBN-LSSVR is put forward. The developed model is valued in terms of the mean absolute error (MAE), the mean absolute percentage error (MAPE), the root mean square error (RMSE), and the coefficient of determination ( R 2 ). Results illustrate that the model proposed in this paper can accurately predict water quality parameters and better robustness of water quality parameters compared with the traditional back propagation (BP) neural network, LSSVR, the DBN neural network, and the DBN-LSSVR combined model.

Cow-calf operations form the foundation of the U.S. beef industry, with approximately 27.9 million beef cows and 25.1 million calves born in 2024, comprising 61.1% of total cattle inventory. Though their exact economic contribution is hard to predict due to the integrated nature of the beef supply chain, the cattle industry contributed approximately $88.4 billion (17%) to the total U.S. agricultural receipts in 2023. Despite their significance, these operations face economic challenges, including fluctuating feed costs, market volatility, and drought. Given the narrow profit margins of these operations, predictive modeling of profitability can offer valuable insights for sustainable management and decision-making. This study applies three univariate (Auto-Regressive Integrated Moving Average (ARIMA), Seasonal ARIMA (SARIMA), SARIMA with Exogenous regressors (SARIMAX)), three multivariate machine learning models (Random Forest (RF), AdaBoost, Support Vector Regression (SVR)), and one deep learning model (Long Short-Term Memory, LSTM) to assess the economic performance of cow-calf operations, specifically predicting the average farm price received for calves ($/Cwt). The analysis leverages historical data along with key domain-specific variables, including feed costs (monthly average U.S. farm price of corn, $/bushel), fuel prices (industrial natural gas price, $/thousand cubic feet), foreign exchange rates (U.S.-Japan exchange rate), and inflation indices (Consumer Price Index, CPI), to provide robust and enhanced predictive insights. A cow-calf budget dataset spanning from 1996 to 2024 was compiled from sources including the USDA Agricultural Marketing Service and Texas A&M AgriLife Extension. Data acquisition was facilitated through web scraping techniques, enabling the automated extraction of the historical economic indicators. The collected data underwent preprocessing and feature engineering, to enhance model input quality, and both univariate time-series and multivariate machine learning models were applied to capture temporal and nonlinear patterns in calf prices. Model performance was evaluated using multiple metrics, including Mean Absolute Error (MAE), Mean Percentage Error (MPE), Root Mean Square Error (RMSE), Root Mean Square Percentage Error (RMSPE), and Coefficient of Determination (R²) to comprehensively assess predictive performance. The analysis revealed substantial relationships between selling price of cow-calf operations and key economic indicators. Among the evaluated forecasting models, SARIMA exhibited the strongest performance by effectively capturing seasonal trends, while SARIMAX showed comparable accuracy through the integration of external variables. Among the machine learning techniques, AdaBoost performed best, outperforming RF and SVR. The LSTM model achieved the overall best performance, demonstrating strong capability in modeling both nonlinear dependencies and temporal dynamics in calf price trends. These findings demonstrate the potential of combining classical time-series and modern machine learning models to support price forecasting in cow-calf operations. This hybrid modeling approach provides a powerful decision-support tool for producers navigating market volatility, and future research should explore real-time forecasting and broader economic and environmental integrations to further strengthen predictive capabilities.

. This study introduces support vector regression (SVR) approach to model the relationship between the glass transition temperature (Tg) and multipole moments for polymers. SVR was trained and tested via 60 samples by using two quantum chemical descriptors including the molecular traceless quadrupole moment and the molecular average hexadecapole moment Φ. The prediction performance of SVR was compared with that of reported quantitative structure property relationship (QSPR) model. The results show that the mean absolute error (MAE), mean absolute percentage error (MAPE) and root mean square error (RMSE) of training samples and test samples achieved by SVR model, are smaller than those achieved by the QSPR model, respectively. This investigation reveals that SVR-based modeling is a practically useful tool in prediction of the glass transition temperature of polymers.

ABSTRACTAboveground biomass (AGB) of mangrove forest plays a crucial role in global carbon cycle by reducing greenhouse gas emissions and mitigating climate change impacts. Monitoring mangrove forests biomass accurately still remains challenging compared to other forest ecosystems. We investigated the usability of machine learning techniques for the estimation of AGB of mangrove plantation at a coastal area of Hai Phong city (Vietnam). The study employed a GIS database and support vector regression (SVR) to build and verify a model of AGB, drawing upon data from a survey in 25 sampling plots and an integration of Advanced Land Observing Satellite-2 Phased Array Type L-band Synthetic Aperture Radar-2 (ALOS-2 PALSAR-2) dual-polarization horizontal transmitting and horizontal receiving (HH) and horizontal transmitting and vertical receiving (HV) and Sentinel-2A multispectral data. The performance of the model was assessed using root mean square error (RMSE), mean absolute error (MAE), coefficient of determination (R2), and leave-one-out cross-validation. Usability of the SVR model was assessed by comparing with four state-of-the-art machine learning techniques, i.e. radial basis function neural networks, multi-layer perceptron neural networks, Gaussian process, and random forest. The SVR model shows a satisfactory result (R2 = 0.596, RMSE = 0.187, MAE = 0.123) and outperforms the four machine learning models. The SVR model-estimated AGB ranged between 36.22 and 230.14 Mg ha−1 (average = 87.67 Mg ha−1). We conclude that an integration of ALOS-2 PALSAR-2 and Sentinel-2A data used with SVR model can improve the AGB accuracy estimation of mangrove plantations in tropical areas.

In recent times, gold has been one of the prioritized commodities in terms of long term as well as short term investments since the investors consider gold as a hedgerow against the unforeseen events leading to chaos in the market. Consequentially, the price of gold in the market plays an important role. In this research work, time-series gold price prediction models have been developed using the support vector regression and anfis models for the prediction of daily gold prices. The support vector model was designed using epsilon support vector regression method while the adaptive neural fuzzy inference systems have been developed using grid partition and subtractive clustering methods. The gold prices obtained for the training and testing were obtained from Perth Mint of Australia. The evaluation criteria for the comparison of the models are Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Nash-Sutcliffe model efficiency coefficient (E) and Mean absolute percentage error (MAPE). It was observed that the models obtained using support vector regression outperformed the ANFIS models. In the ANFIS models, it was observed that ANFIS-GP performed slightly better than the ANFIS-SC model.

Super learner ensemble model: A novel approach for predicting monthly copper price in future

Drought is considered one of the costliest natural disasters that result in water scarcity and crop damage almost every year. Drought monitoring and forecasting are essential for the efficient management of water resources and sustainability in agriculture. However, the design of a consistent drought prediction model based on the dynamic relationship of the drought index with its antecedent values remains a challenging task. In the present research, the SVR (support vector regression) model was hybridized with two different optimization algorithms namely; Particle Swarm Optimization (PSO) and Harris Hawks Optimization (HHO) for reliable prediction of effective drought index (EDI) 1 month ahead, at different locations of Uttarakhand State of India. The inputs of the models were selected through partial autocorrelation function (PACF) analysis. The output produced by the SVR-HHO and SVR-PSO models was compared with the EDI estimated from observed data using five statistical indicators, i.e., RMSE (Root Mean Square Error), MAE (Mean Absolute Error), COC (Coefficient of Correlation), NSE (Nash-Sutcliffe Efficiency), WI (Willmott Index), and graphical inspection of radar-chart, time-variation plot, box-whisker plot, and Taylor diagram. Appraisal of results indicates that the SVR-HHO model (RMSE = 0.535-0.965, MAE = 0.363-0.622, NSE = 0.558-0.860, COC = 0.760-0.930, and WI = 0.862-0.959) outperformed the SVR-PSO model (RMSE = 0.546-0.967, MAE = 0.372-0.625, NSE = 0.556-0.855, COC = 0.758-0.929, and WI = 0.861-0.956) in predicting EDI. Visual inspection of model performances also showed a better performance of SVR-HHO compared to SVR-PSO in replicating the median, inter-quartile range, spread, and pattern of the EDI estimated from observed rainfall. The results indicate that the hybrid SVR-HHO approach can be utilized for reliable EDI predictions in the study area.

Travel Time Prediction (TTP) has become an essential service that people use in daily commutes. With the precise TTP, individuals, logistic companies, and transport authorities can better manage their activities and operations. This paper presents a novel Hybridized Deep Feature Space (HDFS) based TTP ensemble model (HDFS-TTP) for accurate travel time prediction. In the first step, extensive endogenous and exogenous data sources are augmented with traffic data obtained using sensors. Next, we used Principal Component Analysis (PCA) and Deep Stacked Auto-Encoder (DSAE) for feature reduction. We generated feature spaces of deep learning models, namely Convolutional Neural Network (CNN), Multi-Layer Perceptron (MLP), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU), and fed them to a model based on Support Vector Regressor (SVR) for predicting travel times. Two best-performing models are selected, and their feature spaces are hybridized to boost feature space. On this boosted feature space, we employed SVR for final prediction. Our proposed HDFS-TTP ensemble can learn complex non-linearities in traffic data with the varying architectural design. The performance of our proposed HDFS-TTP ensemble using hybridized and boosted feature spaces showed significant improvement in test data in terms of Root Mean Square Error (62.27± 1.58), Mean Absolute Error (13.38± 1.09), Maximum Absolute Error (104.66± 2.77), Mean Absolute Percentage Error (2.50± 0.03), and Coefficient of determination (0.99714± 0.00044).