ABSTRACT Short‐term temperature forecasts are useful in many IoT systems, but the models behind them are usually tested on one large public dataset rather than on the cheap sensors actually deployed. We ran one forecasting pipeline—a naive “no change” baseline and three machine‐learning models (Ridge regression, Random Forest, and XGBoost) over lagged and rolling features—on two sources: the Jena Climate benchmark (Germany, 2009–2016, about 420 000 records) and 1 month of data from each of three low‐cost optical particle‐counter nodes in Tetovo, North Macedonia. All splits were chronological, and every evaluation was repeated with walk‐forward cross‐validation. Two of the three field nodes came close to the benchmark in absolute error (RMSE 0.76°C and 0.78°C, against 0.63°C on Jena) on roughly one hundredth of the training data, and two exported two‐feature models ran in real time on an Arduino MKR1000. The rest of the results were less comfortable. At one of those two nodes, no trained model beat the naive baseline, and walk‐forward evaluation showed the tree ensembles there to be far more unstable than the single split suggested. The third node's much larger error was not a modeling failure: its housing was heating in the sun, which we traced in the raw files and the site operators confirmed. Two further data‐quality defects, one per dataset, each changed a reported error figure. On this evidence, a benchmark score on its own says rather little about how a forecasting pipeline will behave on cheap edge hardware in the field.
ABSTRACT There is a tremendous rise in energy usage when smart infrastructure systems are being rolled out, hence the need to develop sustainable computing and communication technologies that will ensure the proper usage of the available resources. To provide real‐time energy management in smart infrastructure landscape, this paper suggests an energy optimization framework, based on a digital twin, which involves Internet of Things (IoT), edge computing and intelligent modeling. It uses the UCI Energy Efficiency Dataset to simulate the behavior of buildings in terms of energy consumption and builds a data‐driven digital twin to optimize the building behavior dynamically. The suggested strategy will decrease the overall power usage by 245.6 to 168.3 kWh which will result in an enhancement of 31.5%. Also, the overall accuracy of prediction is increased, the root mean square error (RMSE) is decreased to 3.87 and the mean absolute error (MAE) is decreased to 2.76 that are improved by 39.7 or 44.2, respectively. The architecture can be shown to perform efficiently with real‐time performance with processing latency of 5–8 ms and end‐to‐end latency of 12–18 ms, and less communication overheads of 6–9 times through edge‐based processing. These findings embrace the usefulness of the suggested digital twin‐based sustainable computing model in enhancing energy conservation, lowering operational expenses by enabling scalable systems of smart infrastructure.
ABSTRACT Industrialization is growing as a threat to the quality of water in urban lakes. In this paper, a hybrid machine learning system (combining remote sensing [RS] data with a stacked ensemble model) is used to predict the turbidity of Saki Lake within the Patancheruvu industrial belt in Hyderabad. As a result of the nonlinear relationship between RS spectral characteristics and in situ turbidity (NTU), conventional models are ineffective. XGBR + NN + GBR as base learners with Ridge Regression as meta‐learner minimally decreased MAE to 3.06 NTU (compared to 5.11, 4.17, and 3.81 NTU as un‐minimized MAE of individual models). An additional error‐correction step following stacking with the XGBR was used to further decrease MAE to 2.18 NTU ( R 2 = 0.86). The two‐step framework shows that ensemble learning with specific fault correction can provide scaled water quality monitoring applicable to several contaminated lentic water bodies.
ABSTRACT The fast expansion of artificial intelligence (AI) technologies across the ecommerce ecosystems has radically changed the processes of how firms interact with consumers. The paper explores the complex effect of AI‐based personalization on consumer decision making and the firm‐level performance outcomes, based on e‐commerce sites. Our proposed Personalization Impact Optimization Framework (PIOF) represents a blend of collaborative filtering, deep neural networks, and reinforcement learning as a framework to model the user preference dynamics in real‐time. The mathematical model integrates a multi‐objective optimization criterion taking into consideration both consumer utility and revenue of the firm. We show that conversion rates increase with AI‐based engines of personalization by a maximum of 34.7%, consumer search costs with AI‐based personalization engines decrease by 28.3% and the average value of orders with AI‐based personalization engines by 19.6%, compared to controls without personalization. Moreover, our econometric model shows that there is a statistically significant positive correlation between the extent of personalization and the amount of gross merchandise volume (GMV) by the firm. The combination of reinforcement learning and fairness‐aware multi‐objective optimization led to better performance than sequential recommenders based on transformers. Such results have significant implications on platform strategy, algorithmic governance and the consumer welfare policy.
