Indonesia, an archipelagic nation with more than 17,000 islands, faces persistent challenges in providing reliable electricity to remote communities. Decentralized hydropower offers a sustainable solution by utilizing abundant local water resources. Pico-hydropower and micro-hydropower systems provide cost-effective alternatives for rural electrification while supporting sustainable watershed resource utilization. In this study, a comprehensive identification framework was developed to evaluate the feasibility of integrating pico-hydropower with existing micro-hydropower systems to support community-based ice crystal production microenterprises in Indonesia. Candidate sites were selected using a purposive sampling strategy based on flow continuity, hydraulic head ($\Delta H$), discharge ($Q$), accessibility, and installation suitability. Spatial identification was performed through the integration of Google Earth imagery, unmanned aerial vehicle surveys, and total station measurements, followed by systematic field verification. Repeated field measurements were conducted to reduce measurement uncertainty and improve the reliability and reproducibility of the proposed methodology. Laboratory experiments and field validation were subsequently performed using a pico-hydropower portable turbine equipped with an undershot waterwheel configuration. A total of 60 potential installation sites were identified. Laboratory testing demonstrated that the prototype generated output under flow rates ranging from 11 to 25 L s$^{-1}$, producing sufficient electrical power for 50–100 W light-emitting diode lighting applications. Field validation at 30 representative sites confirmed the technical feasibility of the proposed system. The highest performance was observed at P21, where a rotational speed of 1,543 revolutions per minute and an output voltage of 3.1 V were achieved, providing electrical power equivalent to approximately 100 W under the prototype configuration, whereas the minimum validated performance was recorded at P15 under a discharge of 16 L s$^{-1}$. The results demonstrate that distributed pico-hydropower installations can effectively complement existing micro-hydropower infrastructure, supporting productive applications such as ice crystal production. The proposed framework provides a practical and scalable approach for expanding sustainable decentralized hydropower systems in geographically dispersed regions.

Accurate classification of road level of service (LOS) is essential for evaluating traffic operating conditions, supporting road safety assessment, and informing maintenance and rehabilitation planning. In this study, vehicle-count data were collected at the entrance to the car park of Hacettepe University’s Sıhhiye Campus and used to develop machine learning (ML)-based models for LOS classification. Four supervised classification algorithms—K-nearest neighbours (KNN), support vector classification (SVC), AdaBoost, and random forest (RF)—were evaluated using the number of cars, number of trucks, hourly traffic volume, and flow rate as predictor variables. The dataset was divided into training and testing subsets using an 80:20 ratio. To improve model generalisation and reduce the risk of overfitting, five-fold cross-validation combined with grid-search-based hyperparameter optimisation was performed using the training data. Model performance was assessed using accuracy, Cohen’s kappa coefficient, F1 score, recall, precision, confusion matrices, and receiver operating characteristic (ROC) curve analysis. Among the evaluated classifiers, KNN achieved the highest overall performance in LOS classification. To enhance the interpretability of the resulting predictions, Shapley Additive Explanations (SHAP) analysis was subsequently applied to the best-performing model. The SHAP results indicated that flow rate was the most influential predictor of LOS classification, whereas the number of heavy vehicles exerted the smallest overall influence among the variables considered. These findings demonstrate the potential of interpretable ML approaches to complement conventional traffic-performance assessment by providing both predictive classification and insight into the relative contribution of traffic-flow variables to LOS determination.