This paper aims to create an innovative approach to improving IoT-based smart parking systems by integrating machine learning (ML) and Artificial Intelligence (AI) with mathematical approaches in order to increase the accuracy of the parking availability predictions. Three regression-based ML models, random forest, gradient boosting, and LightGBM, were developed and their predictive capability was compared using data collected from three parking locations in Skopje, North Macedonia from 2019 to 2021. The main novelty of this study is based on the use of autoregressive modeling strategies with lagged features and Z-score normalization to improve the accuracy of regression-based time series forecasts. Bayesian optimization was chosen for its ability to efficiently explore the hyperparameter space while minimizing RMSE. The lagged features were able to capture the temporal dependencies more effectively than the other models, resulting in lower RMSE values. The LightGBM model with lagged data produced an R2 of 0.9742 and an RMSE of 0.1580, making it the best model for time series prediction. Furthermore, an IoT-based system architecture was also developed and deployed which included real-time data collection from sensors placed at the entry and exit of the parking lots and from individual slots. The integration of ML, AI, and IoT technologies improves the efficiency of the parking management system, reduces traffic congestion and, most importantly, offers a scalable approach to the development of urban mobility solutions.
The integration of Internet of Things (IoT) technology has profoundly transformed urban life, particularly in the realm of parking management. Smart parking systems harness the capabilities of IoT to optimize parking space utilization, alleviate congestion, and elevate user experience. This chapter delves into the intricate process of data collection within IoT-enabled smart parking environments, with a specific emphasis on the seamless integration of machine learning and artificial intelligence (AI) techniques. By conducting a comprehensive analysis of various data sources, machine learning algorithms, and AI technologies, this chapter elucidates how smart parking systems leverage intelligent data collection and analysis to enhance operational efficiency and effectiveness. Through the convergence of IoT, machine learning, and AI, smart parking systems are poised to revolutionize urban mobility and drive sustainable urban development.
Feature importance analysis is essential for interpreting machine learning models in diabetes mellitus (DM) risk prediction; however, existing interpretability methods often produce inconsistent feature rankings across models. This study proposes a unified ODE-inspired interpretability framework and an algorithmic decision procedure for robust feature selection by integrating contribution-based (SHAP), perturbation-based (permutation importance), and sensitivity-based feature importance measures. Multiple supervised machine learning models, including Logistic Regression, Random Forest, Gradient Boosting, Histogram Gradient Boosting, and Multilayer Perceptron, were trained and evaluated on a longitudinal biochemical and demographic dataset comprising 200 patients with three repeated visits (N = 600 observations). To preserve longitudinal integrity and avoid patient-level information leakage, grouped cross-validation was applied. A sensitivity-based feature importance formulation using finite-difference approximations enabled model-agnostic comparison across heterogeneous machine learning architectures. Stability, normalization, and cross-method agreement analyses were additionally introduced to evaluate consistency of feature rankings across models and interpretability methods. Experimental results consistently identified HbA1c as the dominant predictor, followed by lipid-related variables, age, and body mass index. Strong agreement was observed between ODE-inspired feature importance and SHAP analysis, whereas permutation importance demonstrated comparatively weaker agreement with sensitivity-based methods. The proposed framework further enabled systematic analysis of ranking stability, cross-method agreement, longitudinal sensitivity dynamics, and the introduction of an agreement-weighted Consensus Interpretability Score (CIS) for unified feature ranking across heterogeneous interpretability methods. The results demonstrate that integrating ODE-inspired sensitivity analysis with machine learning provides a robust, interpretable, and computationally scalable framework for feature importance assessment in diabetes risk prediction. The proposed approach offers a principled solution to inconsistent feature importance estimation and supports more reliable interpretation of biomedical machine learning models.
Uncertainty-aware decision-making increasingly relies on multimodal sensing pipelines that must fuse correlated measurements, propagate uncertainty, and trigger reliable control actions. This study develops a unified mathematical framework for multimodal data fusion and Bayesian decision-making under uncertainty. The approach integrates adaptive Covariance Intersection (aCI) for correlation-robust sensor fusion, a Gaussian state–space backbone with Kalman filtering, heteroskedastic Bayesian regression with full posterior sampling via an affine-invariant MCMC sampler, and a Bayesian likelihood-ratio test (LRT) coupled to a risk-sensitive proportional–derivative (PD) control law. Theoretical guarantees are provided by bounding the state covariance under stability conditions, establishing convexity of the aCI weight optimization on the simplex, and deriving a Bayes-risk-optimal decision threshold for the LRT under symmetric Gaussian likelihoods. A proof-of-concept agro-environmental decision-support application is considered, where heterogeneous data streams (IoT soil sensors, meteorological stations, and drone-derived vegetation indices) are fused to generate early-warning alarms for crop stress and to adapt irrigation and fertilization inputs. The proposed pipeline reduces predictive variance and sharpens posterior credible intervals (up to 34% narrower 95% intervals and 44% lower NLL/Brier score under heteroskedastic modeling), while a Bayesian uncertainty-aware controller achieves 14.2% lower water usage and 35.5% fewer false stress alarms compared to a rule-based strategy. The framework is mathematically grounded yet domain-independent, providing a probabilistic pipeline that propagates uncertainty from raw multimodal data to operational control actions, and can be transferred beyond agriculture to robotics, signal processing, and environmental monitoring applications.