Studies have confirmed that PM10, defined as respirable particles with diameters of 10 μm and smaller, has adverse effects on human health and the environment. Various estimation methods are employed to determine the PM10 concentration using historical data on controlling PM10 air pollution, early warning, and protecting public health and the environment. The present study analyses different Long Short-Term Memory (LSTM) models that can predict hourly PM10 concentration. In parallel, the study also investigates the effectiveness of the data preprocessing and feature selection (DPFS) process on the prediction accuracy of the LSTM models. For this purpose, three different LSTM models, namely Vanilla, Bi-Directional, and Stacked, were developed. Then, a comprehensive data preprocessing stage is used to eliminate missing and erroneous data and outliers from real-world raw data, and a feature selection process is applied to extract unnecessary features. The LSTM models consider three air quality parameters, including SO2, O3, and CO, and three meteorological factors, including relative humidity, wind direction, and wind speed. The prediction performances of the LSTM models are compared using the RMSE, MAE and R2 performance index according to whether DPFS is used in the models or not. As a result, when the DPFS process was applied, the proposed LSTM models achieved high prediction performance and can be used to predict hourly PM10 concentrations. Overall, the DPFS process significantly enhanced the developed LSTM models’ prediction performance. Furthermore, the proposed model might be a useful tool for city administrators to make decisions and improve air quality management efforts.

Since air pollution negatively affects human health and causes serious diseases, accurate air pollution prediction is essential regarding environmental sustainability. Although conventional statistical and machine learning methods have been widely used for air quality forecasting, they have limitations in finding nonlinear relations and modeling sequential data. In recent years, deep learning methods such as long short-term memory, recurrent neural networks, and gated recurrent units have been successfully applied in several research areas, including time-series forecasting. In this study, deep learning algorithm is employed to predict the PM2.5 dataset, including air pollutants (NO, NO2, NOX, O3, PM2.5, SO, and SO2) and meteorological features (wind speed, wind direction, and air temperature) in Istanbul metropolitan. Deep learning algorithms have many hyperparameters such as learning and dropout rate, the number of hidden layers and units in each hidden layer, activation function, loss function, and optimizer that need to be optimized in order to achieve optimal training performance. Therefore, a genetic algorithm-based hyperparameter optimization approach is proposed to find the best parameter combination. The prediction results of deep learning algorithms are compared with default hyperparameters and random search algorithms to confirm the efficacy of the genetic algorithm approach. The proposed method outperforms the other configurations, with the MSE error reduced by 13.38% and 55.30% for testing performance, respectively. The experimental results revealed that genetic algorithms are promising and applicable in hyperparameter optimization of deep neural network models, especially in air quality forecasting.

Poor air quality has various detrimental physical and mental effects on human health and quality of life. In particular, PM2.5 air pollution has been associated with cardiovascular and respiratory problems. Therefore, air quality management is an essential issue for densely populated cities to reduce or prevent the adverse effects of air pollution. Considering this, reliable models for predicting pollution levels for pollutants like PM2.5 are critical tools for decision-making. For this purpose, this study presents three kinds of deep learning (DL) algorithms (LSTM, RNN, and GRU) that utilize a time-windowing strategy to predict the hourly concentration of PM2.5 in the Istanbul metropolitan. The models were trained and tested using large data sets that envelope air quality parameters (PM2.5, SO2, NO, NO2, NOX, and O3) and meteorological factors (temperature, wind speed, relative humidity, and air pressure) for about five years. The experimental results demonstrate that the LSTM+LSTM model performs significantly better with an R2 of 0.98 and 0.97 at the significance level (p < 0.05) for training and test sets compared to other deep learning algorithms. In addition, data for one year from several stations located in nine different districts of Istanbul were used to evaluate the proposed model’s generalization ability. As a result, the proposed LSTM+LSTM model has a good generalization ability with an R2 accuracy rate of 0.90 (p < 0.05) and above for all stations and can be used for non-linear, non-stationary multidimensional time series data. Furthermore, the results were compared to other studies in the literature; it was found that the proposed LSTM+LSTM model performed better in predicting PM2.5 concentrations.

Finding optimal solutions to production planning and scheduling problems is crucial for surviving in a competitive environment and meeting customer expectations over time. Planning can become complicated in sectors with many different products such as tire production. This study focuses on the bottleneck problem caused by a machine called a Quadruplex Extruder in a tire factory. With this machine, rubber is extruded and transformed into a tread material product, which is critically important in some essential tire features, such as low rolling resistance and brake distance. This study aims to minimize the set-up times in production by optimizing the manufacturing order of the products produced in a quadruplex extruder machine using the Ant Colony Algorithm (ACA), a well-known metaheuristic method to solve polynomial optimization problems. In addition, the second version of the Lin–Kernighan–Helsgaun (LKH-2) algorithm was adapted to this problem. Manually prepared, LKH-2 and ACA-produced schedules were compared in terms of global efficiency. As a result, it has been shown that ACA can provide fast and suitable solutions for decision makers in production planning.

Soil liquefaction during earthquakes is a complex geotechnical engineering problem. Although various analytical approaches exist for predicting liquefaction risk, their limitations have led researchers to explore using artificial intelligence and machine learning methods. Machine learning has the potential to significantly improve the ability to predict soil liquefaction and mitigate the associated risks. This study proposes a hyper parameterized artificial neural network architecture using random search, grid search, and Bayesian optimization algorithms to predict the factor of safety against liquefaction. The performances of hyper parameterized machine learning algorithms, including artificial neural networks (ANN), decision trees (DT), random forest (RF), and support vector regression (SVR), were compared. Statistical tests show that the proposed ANN outperformed the others with a determination coefficient of 0.99 at a 95% significance level. Hyperparameter optimization significantly improved learning performance with up to a 48% reduction in RMSE scores. The proposed method was compared with previous studies, and performance results confirmed its effectiveness and generalization ability. In conclusion, this study highlights the potential of machine learning algorithms for predicting soil liquefaction and emphasizes the importance of hyperparameter optimization for improving model performance. The findings of this study have practical implications for improving liquefaction risk assessment and mitigating the associated hazards.

Since soil liquefaction is a dimension that increases the amount and severity of losses in an earthquake, it is vital to estimate the liquefaction potential accurately. Traditionally, several analytical inferences were made for the prediction of soil liquefaction. However, it is necessary to use machine learning methods to establish nonlinear relationships of soil physical characteristics and develop an accurate classification model. In this study, the applicability of seven different machine learning algorithms; decision trees, logistic regression, support vector machines, k-nearest neighbors, stochastic gradient descent, random forest, and artificial neural network, were investigated on a data set obtained from field experiments (Standard Penetration Test) on soils in Adapazari region after the 1999 earthquake. Performance metrics such as accuracy, recall, precision, F1 score, and receiver operating characteristic evaluated algorithms. As a result of experimental studies, the decision tree algorithm performed best on the dataset, with an overall accuracy of 90%. The decision tree model provides an easy and effective tool for evaluating ground liquefaction potential to decision-makers. As a result of the decision tree study, it was observed that the mean grain size (D50) soil feature has the most significant effect on the liquefaction potential.