Statistical Metrics Research Articles

Long Short Term Memory Based Traffic Prediction Using Multi-Source Data

AbstractTraffic prediction is a task where the goal is to determine the number and type of vehicles, or some other traffic related metric, at certain time point. In addition to predicting the short-term evolution of traffic, prediction can be done for estimating traffic for distant future based on the trends found in historical traffic data, which is a critical component of traffic simulators being able to spawn realistic number of vehicles under prevailing situation. Such prediction system needs to be dependent on the characteristics of the situation and not the preceding traffic flow. This work presents a deep learning based prediction pipeline that uses a Long Short Term Memory (LSTM) network to map temporal, weather and traffic accident data accurately into traffic flow to predict traffic flow over multiple timesteps from various non-traffic inputs. Traffic data can then be produced based on independent data like weather forecasts and be used for other applications. As far as we know, no previous traffic predictor combines so many input variables to predict traffic flow with vehicle type information. To make the event based traffic accident dataset compatible with time series data, a novel preprocessing step based on power law decay phenomenon is added. Quantitative experiments show that the proposed preprocessing step and optimized hyperparameters improve the accuracy of the predictor on multiple metrics compared to a model without accident information. In two established statistical evaluation metrics, Mean Absolute Error and Mean Squared Error, the improvement was over $$20 \%$$ 20 % for certain vehicle types.

Modeling nonlinear behavior of piezoelectric actuators using improved WT-GRU neural network

The coupling of hysteresis and low damping vibration in a piezoelectric actuator results in low modeling accuracy and adversely affects output motion precision. To mitigate this problem, this paper proposes an improved wavelet transform gated recurrent unit (WT-GRU) model. This model integrates a convolutional layer with WT-GRU to enhance its capability to express complex relationships. Firstly, the input voltage sequence undergoes wavelet transform to decompose it into a set of frequency subsequences. Then, using the feature extraction and representation ability of the convolutional layer, significant features are extracted from subsequences to construct a time series feature vector. Finally, a gated recurrent unit is trained to predict the output displacement sequence accurately. Statistical metrics such as mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination ( R2) were utilized to evaluate the model performance. Additionally, experimental tests were conducted with frequency excitation signals at 80–120 Hz. Experimental results show that the proposed model achieves a MAE of 0.0102 mm, an RMSE of 0.0148 mm, and an R2 value of 0.9478. This model exhibits a significant advantage in accurately predicting the output displacement of piezoelectric actuators, thereby providing a reliable foundation for designing piezoelectric actuator control systems.

Rainfall From Brazilian Flying Rivers: Evaluating the Effectiveness of Precipitation Gridded Databases

ABSTRACTThe uneven global distribution of rainfall significantly impacts water resources and environmental sustainability, emphasising the need for reliable climate prediction models. Accurate predictions are vital for sectors such as food security, urban planning and disaster management. Data from ground stations, radars and satellites are essential, despite challenges like instrumental errors. Satellites, with their comprehensive sensors, are crucial for atmospheric observations, aiding in the prediction of large‐scale climatic events. Climate models such as CHIRPS, GLDAS, TerraClimate, and PERSIANN use different approaches to analyse precipitation data, which is key to understanding its spatial and temporal variability. This study evaluated (rainfall data) from these four climate models over 20 years (within the Brazilian territory), focusing on the spatiotemporal behaviour of rainfall using statistical metrics such as R2, RMSE, and MAPE. The findings showed that CHIRPS had the best performance (R2 = 0.843; RMSE = 42.83; MAPE = 0.09%), excelling in both overall database and extreme event analyses. TerraClimate, initially the lowest‐performing model (R2 = 0.413; RMSE = 91.56; MAPE = 0.23%), improved significantly when combined with elevation through multiple linear regression (MLR), achieving R2 of 0.718, RMSE of 31.14, and MAPE of 9.56%. This made TerraClimate a viable model for studying the Flying Rivers. The study highlights that model selection should align with the specific characteristics of the area under consideration, with CHIRPS being particularly suitable for the studied region. This research enhances the understanding of the effectiveness of these models in estimating rainfall compared to in situ measurements, which is crucial for various applications. The authors advocate for further studies to advance research on the Flying Rivers, their significance, and the impacts of climate change on them.

An interpretable and explainable deep learning model for predicting hydrogen solubility in diverse chemicals

In this study, an explainable and interpretable deep learning (DL) model based on convolutional neural network (CNN) was suggested to accurately estimate H2 solubility in various chemicals under vast ranges of pressure and temperature. The model was implemented using more than 3700 authenticated datapoints. The results revealed that the CNN model achieved excellent predictions and surpassed the well-known machine learning (ML) and prior predictive paradigms. In this context, the CNN demonstrated attractive statistical metrics (RMSE = 0.0049 and R2 = 0.9934). The explainability and interpretability of the suggested DL-based model were testified using the Shapley Additive exPlanations (SHAP) method. Additionally, trend analyses were conducted on the model’s predictions to verify that it accurately reflects H2 solubility trends in various chemicals at different pressure and temperature levels. Lastly, the capability of the introduced DL model greatly improves the simulation of processes involving this crucial parameter in both industrial and academic sectors.

A designed predictive modelling strategy based on data decomposition and machine learning to forecast solar radiation

Consistent with the United Nations Sustainable Development Goal 7, the design and optimization of clean energy resources (e.g., solar energy) is a highly motivating task for all researchers globally who continue to build research synergies that can tackle the future likelihood of energy crises due to socioeconomic and strategic environmental policy. In this paper, a weekly solar radiation (SR) forecasting model is designed using a robust local mean decomposition (RLMD) technique unified with a random forest (RF) algorithm to generate a fully optimised hybridized RLD-RF model that has a promising capability to forecast the SR values. In the first stage of model design, the RLMD, a frequency resolution method, is applied to decompose the original SR time series into amplitude modulation subseries (AMs), frequency modulation subseries (FMs), and the low-frequency product functions (PFs) to reveal the internal structure of the model construction data to incrementally optimize the RLD-RF model where only PFs were used. Subsequently, the statistically significant lagged subseries at a week ahead forecasting horizon (t – 1) of the low-frequency PFs with residual components are extracted individually, via partial autocorrelation function (PACF), to capture the historical behaviour of frequency-resolved SR component in order to build a robust modelling framework. Consequently, the random forest (RF) algorithm is employed to forecast each of the subseries using PACF-based lagged inputs to construct a fully optimised hybrid RLMD-RF predictive model. RLMD-RF is benchmarked against a baseline RF, M5tree, and multiple linear regression (MLR), Artificial neural network (ANN) and Gaussian process regression (GPR) algorithms, including their hybridized counterparts (i.e., RLMD-M5tree, RLMD-MLR, RLMD-ANN, and RLMD-GPR) using statistical score metrics in the independent testing phase. The results generated at test sites in Queensland State, Australia that have high solar energy potential confirm that the RLMD-RF method can produce quality predictions of weekly solar radiation against the benchmarking comparison models. For instance, RLMD-RF for Barcaldine are higher in terms (EWI= 0.938, ENS= 0.878) against RLMD-MLR (EWI = 0.845, ENS = 0.705), the RLMD-M5tree (EWI = 0.836, ENS = 0.684), RLMD-ANN (EWI = 0.836, ENS = 0.715), RLMD-GPR (EWI = 0.839, ENS = 0.716), the RF (EWI = 0.720, ENS = 0.564), the M5tree (EWI = 0.692, ENS = 0.522), the MLR (EWI = 0.683, ENS = 0.508), the ANN (EWI = 0.708, ENS = 0.519) and the GPR (EWI = 0.708, ENS = 0.520). Similarly, the RLMD-RF also outperformed in Rockhampton, Clermont, and Lockyer Valley stations as compared to other models. This research establishes the practical usefulness of hybridised RLMD-RF modelling framework for accurate SR forecasting and advocates its possible consideration in renewable and sustainable energy production and monitoring systems that can aid in decision-making by energy utilities and stakeholders (e.g., climate and energy policy experts).

Machine-Learning-Based Validation of Microsoft Azure Kinect in Measuring Gait Profiles

Gait is one of the most extensively studied motor tasks using motion capture systems, the gold standard for instrumental gait analysis. Various sensor-based solutions have been recently proposed to evaluate gait parameters, typically providing lower accuracy but greater flexibility. Validation procedures are crucial to assess the measurement accuracy of these solutions since residual errors may arise from environmental, methodological, or processing factors. This study aims to enhance validation by employing machine learning techniques to investigate the impact of such errors on the overall assessment of gait profiles. Two datasets of gait trials, collected from healthy and post-stroke subjects using a motion capture system and a 3D camera-based system, were considered. The estimated gait profiles include spatiotemporal, asymmetry, and body center of mass parameters to capture various normal and pathologic gait peculiarities. Machine learning models show the equivalence and the high level of agreement and concordance between the measurement systems in assessing gait profiles (accuracy: 98.7%). In addition, they demonstrate data interchangeability and integrability despite residual errors identified by traditional statistical metrics. These findings suggest that validation procedures can extend beyond strict measurement differences to comprehensively assess gait performance.

Autonomous Robot Goal Seeking and Collision Avoidance in the Physical World: An Automated Learning and Evaluation Framework Based on the PPO Method

Mobile robot navigation is a critical aspect of robotics, with applications spanning from service robots to industrial automation. However, navigating in complex and dynamic environments poses many challenges, such as avoiding obstacles, making decisions in real-time, and adapting to new situations. Reinforcement Learning (RL) has emerged as a promising approach to enable robots to learn navigation policies from their interactions with the environment. However, application of RL methods to real-world tasks such as mobile robot navigation, and evaluating their performance under various training–testing settings has not been sufficiently researched. In this paper, we have designed an evaluation framework that investigates the RL algorithm’s generalization capability in regard to unseen scenarios in terms of learning convergence and success rates by transferring learned policies in simulation to physical environments. To achieve this, we designed a simulated environment in Gazebo for training the robot over a high number of episodes. The training environment closely mimics the typical indoor scenarios that a mobile robot can encounter, replicating real-world challenges. For evaluation, we designed physical environments with and without unforeseen indoor scenarios. This evaluation framework outputs statistical metrics, which we then use to conduct an extensive study on a deep RL method, namely the proximal policy optimization (PPO). The results provide valuable insights into the strengths and limitations of the method for mobile robot navigation. Our experiments demonstrate that the trained model from simulations can be deployed to the previously unseen physical world with a success rate of over 88%. The insights gained from our study can assist practitioners and researchers in selecting suitable RL approaches and training–testing settings for their specific robotic navigation tasks.

Accurate Method for Solar Power Generation Estimation for Different PV (Photovoltaic Panels) Technologies

In 2023, solar photovoltaic energy alone accounted for 75% of the global increase in renewable capacity. Moreover, this natural energy resource is the one that requires the least investment, which makes it accessible to developing countries. Increasing return on investment in these regions requires a particular evaluation of environmental parameters influencing PV systems performance. Higher temperatures decrease PV module efficiency and, as a result, their power output. Additionally, fluctuations in solar irradiance directly impact the energy generated by these systems. Consequently, it is essential for investors to improve accurate predictive models that assess the power generation capacity of photovoltaic systems under local environmental conditions. Therefore, accurate estimation of maximum power generation is then crucial for optimizing photovoltaic (PV) system performances and selecting suitable PV modules for specific climates. In this context, this study presents an experimental comparison of three maximum power prediction methods for four PV module types (amorphous silicon, monocrystalline silicon, micromorphous silicon, and polycrystalline silicon) under real outdoor conditions. Experimental data gathered over the course of a year are analyzed and processed for the four PV technologies. Three different methods taking into account environmental parameters are presented and analyzed. The first estimation method utilizes irradiance as the primary input parameter, while two additional methods incorporate ambient temperature and PV module temperature for enhanced accuracy. The performance of each method is evaluated using standard statistical metrics, including the root mean square error (RMSE) and coefficient of determination (R2). The results demonstrate the effectiveness of all three methods, with RMSE values ranging from 1.6 W to 3.8 W and R2 values consistently above 0.95. The most appropriate method for estimating PV power output is determined by the specific type of photovoltaic module and the availability of meteorological parameters. This study provides valuable insights for selecting an appropriate maximum power prediction method and choosing the most suitable PV module for a given climate.

Impact of WRF Model Parameterization Settings on the Quality of Short-Term Weather Forecasts over Poland

This research examines the impact of various parameterization settings within the Weather Research and Forecasting (WRF) model on the accuracy of short-term weather forecasts for Poland. The study focuses on the sensitivity of key meteorological variables—namely, air temperature, wind speed, relative humidity, and atmospheric pressure—to different combinations of physical parameterization schemes. Utilizing data from the Global Forecast System (GFS) spanning 2019 to 2022, a series of model simulations were conducted with support from the Poznań Supercomputing and Networking Center (PCSS). To assess the model’s performance across different weather stations, statistical metrics such as the mean absolute error (MAE) and root mean square error (RMSE) were employed. The findings indicate that the configuration labeled “p2” produced the most accurate forecasts for temperature, wind speed, and atmospheric pressure, achieving MAE values of 1.5 °C, 1.6 m/s, and 2 hPa, respectively. However, forecast inaccuracies were notably higher in mountainous regions, particularly regarding wind speed. These results underscore the importance of selecting appropriate parameterization settings tailored to regional characteristics, as different configurations can significantly impact the forecast accuracy, especially in complex terrains. This study contributes to the understanding of short-term weather forecasting models for Central Europe, offering potential pathways for improving localized forecast accuracy.

Experimental analysis and gene expression programming optimization of sustainable concrete containing mineral fillers.

Rapid urbanization has led to a high demand for concrete, causing a significant depletion of vital natural resources, notably river sand, which is crucial in the manufacturing process of concrete. As a result, there is a growing need for environmentally sustainable alternatives to fine aggregate in concrete. Quarry dust (QD) has evolved as a viable and ecologically friendly substitute in response to this demand. In the past, limited experimental investigations and only conventional modeling techniques were used to promote sustainable mineral fillers in concrete. This study proposed a robust soft computing technique using gene-expression programming (GEP) to enhance the usability of sustainable alternatives in concrete. Initially, an experimental study was carried out to examine the feasibility and mechanical characteristics of concrete made from materials including quarry dust and superplasticizer as a partial replacement for fine aggregate. Ten mixed proportions with various proportions (0%, 20%, 40%, and 60%) of quarry dust were used to make M15 and M20 grades of concrete. A series of experimental tests, such as workability, compressive strength (CS), and tensile strength (TS), were conducted to examine the fresh and hardened properties of modified concrete. The established database from the experimental investigations was then used to develop machine learning (ML) models using GEP. The outcomes of the GEP models were validated by comparing them with multi-linear regression (MLR) models and using various statistical metrics such as root mean squared error (RMSE), performance index (PI), correlation coefficient (R), and external validation methods. Finally, sensitivity analysis was performed to investigate the influence of ingredients such as mineral fillers, superplasticizers, and others on the mechanical properties of concrete. To enhance the practical usage of the study, a graphical user interface (GUI) was also created. The study revealed that 40% of the replacement of fine aggregates with mineral filler and superplasticizer shows the optimum properties. GEP models outperformed MLR, achieving R² values of 0.96 in CS and 0.92 in TS, compared to MLR's lower values of 0.85 in CS and 0.81 in TS. The proposed GEP equations and user-friendly GUI can be used to develop the pre-mix design of concrete using quarry dust and superplasticizers.

Identification of Anti-Tuberculosis Drugs Targeting DNA Gyrase A and Serine/Threonine Protein Kinase PknB: A Machine Learning-Assisted Drug-Repurposing Approach

Tuberculosis (TB) is a global health challenge associated with considerable levels of illness and mortality worldwide. The development of innovative therapeutic strategies is crucial to combat the rise of drug-resistant TB strains. DNA Gyrase A (GyrA) and serine/threonine protein kinase (PknB) are promising targets for new TB medications. This study employed techniques such as similarity searches, molecular docking analyses, machine learning (ML)-driven absolute binding-free energy calculations, and molecular dynamics (MD) simulations to find potential drug candidates. By combining ligand- and structure-based methods with ML principles and MD simulations, a novel strategy was proposed for identifying small molecules. Drugs with structural similarities to existing TB therapies were assessed for their binding affinity to GyrA and PknB through various docking approaches and ML-based predictions. A detailed analysis identified six promising compounds for each target, such as DB00199, DB01220, DB06827, DB11753, DB14631, and DB14703 for GyrA; and DB00547, DB00615, DB06827, DB14644, DB11753, and DB14703 for PknB. Notably, DB11753 and DB14703 show significant potential for both targets. Furthermore, MD simulations’ statistical metrics confirm the drug–target complexes’ stability, with MM-GBSA analyses underscoring their strong binding affinity, indicating their promise for TB treatment even though they were not initially designed for this disease.

SARIMA Statistical Model to Predict the Consumption of Anxiolytics as a Treatment for Mental Illnesses

The prevalence of mental health diseases and excessive consumption of anxiolytics has increased in the world. In this scenario, the need arises to determine a model that describes the behavior of pharmacological consumption of anxiolytics in Ecuador, in addition to allowing this general behavior to be projected over time. With a descriptive, exploratory, and non-experimental methodological approach conditioned on obtaining statistical data from official national and international organizations. The population of interest was generalized using flow-type temporal data on the effective consumption of anxiolytics, consisting of 144 monthly records in the period from January 2011 to December 2022. The records represent the proportion of people who consume anxiolytics in relation to the population total available in the statistics of community health care with mental illness disorders of the Ministry of Public Health. In this sense, a viable option is the construction of a temporary SARIMA model. Due to its temporal nature and the management of monthly records, robust estimation was chosen as an option by applying machine learning that efficiently decomposes and extracts both the seasonal and trend components present in the data. Determining the pharmacological consumption of anxiolytics depends on the seasonal factor (months) and the presence of a marked tendency to gradually increase over time, a situation that must be regulated because it represents a situation of drug dependence and overdose. Furthermore, the built model presented adequate suitability when quantifying statistical metrics: RMSE = 5.25% and MAPE = 1%. It is concluded that the proposed model explains the behavior of the consumption of anxiolytics in Ecuador to mitigate situations that occurred in the affected person (anxiety or depression) in the last three months, according to the specification of deterministic and random components identified in the estimated model.

Modeling farmer behavior in adopting agroecological practices using AI tools in shea butter agroforestry systems

This study examines the performance of machine learning algorithms for identifying importance features for agroecological practices adoption in shea agroforestry systems. Primary data were collected from 272 representative and randomly selected farmers in two regions of northern Benin. Four machine learning algorithms (Naïve Bayes, Neural Network, Support Vector Machine and Bagging Decision Trees) were compared using four statistical performance metrics: accuracy, balanced accuracy, recall, and the area under the receiver operating characteristic curve (AUC), as well as calibration plots. The results indicated that the Naïve Bayes model performed best for predicting animal parking practices, and biopesticide while the Support Vector Machine outperformed other models in predicting cultural associations, improved seeds, and organic fertilizers adoption. Feature importance analysis revealed that the most significant feature influencing the adoption of animal parking and organic fertilizers practices was the farmers' region of origin, with the highest importance value (100%). When considering features with at least 50% importance, main occupation and means of transport for manure were also relevant for the adoption of animal parking practice. The farmers' ethnic group, village, and age class were the most influential characteristics for biopesticides, cultural association, and improved seeds practices, respectively, each with a 100% importance value. For improved seeds practice, farm assets and secondary occupation were also relevant when considering features with at least 50% importance. Therefore, providing comprehensive training on agroecological practices such as animal parking, crop diversification, biopesticide management, organic fertilization, and improved seeds use to farmers in Northern Benin, particularly those managing shea agrosystems, could contribute to the sustainable development of this vital resource and enhance food security.

Leveraging machine learning to model salinity and water flux for improved insights into forward osmosis membrane bioreactors

Forward osmosis (FO) and integrated FO with membrane bioreactors (FOMBR), have recently been recognized as highly effective solutions for water treatment and purification. FOMBR operation is intricate due to the combination of membrane filtration and biological treatment, along with the influence of various parameters. To improve the understanding of FOMBR for its potential applications, an accurate prediction model is crucial for enhancing performance and optimizing parameters. Therefore, this study employed various machine learning (ML) approaches to estimate the water flux (Jw) and total dissolved solids in a lab-scale FOMBR system. An extensive dataset was collected from the literature for model development. Different statistical metrics were used to assess the efficacy of the developed models. Two different explainability approaches, partial dependence plots (PDP) and SHapley Additive exPlanation (SHAP), were used to make the predictions of the ML models more explainable and transparent. Notably, the gradient boosting (GB) algorithm achieved an excellent R score of 0.99, while decision tree (DT) and support vector regression (SVR) achieved lower R scores of 0.9719 and 0.9701, respectively, in estimating water flux. Similarly, the GB model performed exceptionally well in estimating TDS of FOMBR. However, the DT and SVR models exhibited the lowest prediction performance compared to the other developed models in estimating the TDS. Overall, the GB algorithm and SVR integrated with boosting and bagging algorithms showed superior performance in predicting both TDS and water flux. In contrast, the standalone models, DT and SVR, provided poor prediction performance. The SHAP and PDP analyses demonstrated that phosphate level was the primary factor influencing both TDS and water flux.

Genetic Algorithm for Atmospheric Correction (GAAC) of water bodies impacted by adjacency effects

Adjacency effect (AE) corrections over inland water surfaces has been a known issue in space-borne optical remote sensing over more than four decades. Here we present a novel algorithm able to simultaneously retrieve the aerosol optical depth, sun glint, AE, water reflectance, and water inherent optical properties (IOPs). The method was evaluated against an in situ data set of remote sensing reflectance (Rrs) collected in ∼100 lakes across Canada. The new algorithm is based on a genetic optimization scheme (GAAC: Genetic Algorithm for Atmospheric Correction), and was here compared to the most popular atmospheric correction algorithms available (ACOLITE, iCOR+SIMEC). The statistical metrics of the Rrs retrieval were improved by a factor of almost 2 in all wavelengths, and for all metrics (Bias, Error, Similarity Angle) relative to other algorithms. Demonstrations of GAAC on scenes of Lansdat-8 OLI, and Sentinel-2 MSI sensors demonstrate the algorithm’s robustness when applied to spatially complex small lake (∼10 km of width) surfaces.

Phenomics Demonstrates Cytokines Additive Induction of Epithelial to Mesenchymal Transition.

Epithelial to mesenchymal transition (EMT) is highly plastic with a programme where cells lose adhesion and become more motile. EMT heterogeneity is one of the factors for disease progression and chemoresistance in cancer. Omics characterisations are costly and challenging to use. We developed single cell phenomics with easy to use wide-field fluorescence microscopy. We analyse over 70,000 cells and combined 53 features. Our simplistic pipeline allows efficient tracking of EMT plasticity, with a single statistical metric. We discriminate four high EMT plasticity cancer cell lines along the EMT spectrum. We test two cytokines, inducing EMT in all cell lines, alone or in combination. The single cell EMT metrics demonstrate the additive effect of cytokines combination on EMT independently of cell line EMT spectrum. The effects of cytokines are also observed at the front of migration during wound healing assay. Single cell phenomics is uniquely suited to characterise the cellular heterogeneity in response to complex microenvironment and show potential for drug testing assays.

Daily Estimates of Global Radiation in the Brazilian Amazon from Simplified Models

Solar radiation is an element and a meteorological factor that is present in several processes, such as evapotranspiration, photosynthesis, and energy generation, among others. However, in some regions, there is a limitation in surface data measurements. In this study, 87 empirical models were evaluated to estimate global radiation (Hg) in the Brazilian Amazon biome; these models were divided into five groups according to the input variables, such as insolation (group I), air temperature (group II), relative humidity (group III), astronomical variables (group IV), and hybrid models (group V). The estimates were evaluated by their significance (t-test) and then according to the statistical metrics of the models’ performance (R2, MBE, RMSE, and d). The group V model Hg/H0=a+bln∆T+cS/S0d presented the best statistical performance in all the evaluated indicators, followed by the group I model Hg/H0=a+bS/S0c, and then the group II model Hg/H0=a+b∆T+c∆T0.25+d∆T0.5+eTmed/H0. The group III models presented a low statistical performance, and the group IV models were not significant (NS) in all the evaluated meteorological stations. In general, it was found that the simplified or hybrid models based on insolation and air temperature were efficient in estimating Hg in the Brazilian Amazon biome.

Electricity demand uncertainty modeling with Temporal Convolution Neural Network models

This work presents a Temporal Convolution Network (TCN) model for half-hourly, three-hourly and daily-time step to predict electricity demand (G) with associated uncertainties for sites in Southeast Queensland Australia. In addition to multi-step predictions, the TCN model is applied for probabilistic predictions of G where the aleatoric and epistemic uncertainties are quantified using maximum likelihood and Monte Carlo Dropout methodologies. The benchmarks of TCN model include an attention-based, bi-directional, gated recurrent unit, seq2seq, encoder–decoder, recurrent neural networks and natural gradient boosting models. The testing results show that the proposed TCN model attains the lowest relative root mean square error of 5.336-7.547% compared with significantly larger errors for all benchmark models. In respect to the 95% confidence interval using the Diebold–Mariano test statistic and key performance metrics, the proposed TCN model is better than benchmark models, capturing a lower value of total uncertainty, as well as the aleatoric and epistemic uncertainty. The root mean square error and total uncertainty registered for all of the forecast horizons shows that the benchmark models registered relatively larger errors arising from the epistemic uncertainty in predicted electricity demand. The results obtained for TCN, measured by the quality of prediction intervals representing an interval with upper and lower bound errors, registered a greater reliability factor as this model was likely to produce prediction interval that were higher than benchmark models at all prediction intervals. These results demonstrate the effectiveness of TCN approach in electricity demand modelling, and therefore advocates its usefulness in now-casting and forecasting systems.

IMPLEMENTATION OF MULTIPLE LINEAR REGRESSION ALGORITHM IN PREDICTING RED CHILI PRICES IN GARUT REGENCY

Vegetables, including red chili peppers, play an important role in food and economic balance. Significant price fluctuations and inflation are often problems for farmers and traders. Garut Regency, as the center of red chili production in West Java, faces similar challenges. This research aims to implement a Multiple Linear Regression algorithm to predict the price of red chili peppers in the Garut Regency, highlighting the novelty of using a combination of One Hot Encoding, Feature Engineering, Standard Scaler, and Hyperparameter Tuning techniques. The method used is CRISP-DM with 6 stages: Business Understanding, Data Understanding, Data Preparation, Modeling, Evaluation, and Deployment. The data used is the price and production of red chili peppers per week in 2018-2023, with a total of 702 records. This research involved 8 trials with data transformation and normalization scenarios. The model evaluation used MSE, RMSE, MAPE, R-squared, and statistical hypothesis testing metrics. Results showed 5 significantly influential attributes: year, month, production, net harvested area, and productivity. The best model yielded MSE 202,134,650, RMSE 14,217, MAPE 29.16%, and R-squared 0.320. This approach is simpler yet effective and is able to provide fairly accurate predictions. This research is expected to contribute to providing predictive models that help farmers and traders anticipate price fluctuations, as well as provide insights for policymakers in price management.

The old good landscape maps: New interpretations enabling ecosystem services assessment of conservation potential at a national scale

The ecosystem services (ES) approach has been widely accepted in environmental policies and management as an adequate platform that can serve as a link between nature and society. Many ES are influenced by the landscape structure. Thus, national-scale landscape mapping can potentially contribute to nature conservation management. However, there are no attempts to directly link the ES assessment with the landscape units at a national level. In this paper, we propose an approach for the transformation of paper copy information from old landscape maps to enable the assessment of ES conservation potential at the national landscape scale. The conceptual scheme of the approach contains three main elements: (i) data acquisition; (ii) landscape and ES assessment data processing; and (iii) mapping of ES potential at a landscape level. The results reveal the landscape heterogeneity based on landscape classification and mapping at a national level and the ES conservation potential based on the analyses of the Natural Heritage (NH) in the country to provide ES. The assessment of ES conservation potential using the national scale landscape mapping allows us to analyze the spatial relationships between the landscapes with high conservation value and the existing nature protection network. The conceptual scheme of the study demonstrates how the results of the ES potential provided by the NH at a national level can be combined with the landscape units from the traditional landscape classification schemes to produce various spatial and statistical metrics that reveal how the national system of protected areas coincides with the areas of high ES conservation value.