Metaheuristic Optimization Review

Accurate wind power forecasting is essential for reliable renewable energy integration, grid stability, reserve scheduling, and wind farm operation because turbine output is highly variable and strongly influenced by meteorological conditions. However, forecasting wind power remains challenging due to the nonlinear relationship between weather variables and power generation, the temporal dependency of hourly observations, and the circular nature of wind direction data. This study aims to develop and compare deep learning models for predicting normalized wind turbine power output using a field-based hourly dataset collected from an operational wind energy site starting from January 2, 2017. The dataset includes temperature, relative humidity, dew point, wind speed at 10 m and 100 m, wind direction at 10 m and 100 m, wind gusts, and normalized turbine output. Five predictive models, namely LSTM, RNN, GRU, CNN, and Dense neural networks, were trained and evaluated after applying data preprocessing procedures, including missing-value handling, feature scaling, temporal alignment, and wind-direction transformation. Model performance was assessed using MSE, RMSE, MAE, MBE, correlation coefficient (r), coefficient of determination (R2), RRMSE, NSE, and WI. The empirical results showed that recurrent architectures outperformed the CNN and Dense models, confirming the importance of temporal learning in hourly wind power forecasting. Among all models, LSTM achieved the best overall performance, with MSE = 0.0008, RMSE = 0.0282, MAE = 0.0106, MBE = -0.0006, r = 0.9940, R2 = 0.9880, RRMSE = 0.0861, NSE = 0.9880, and WI = 0.9970. These findings demonstrate that LSTM can effectively capture nonlinear and sequential relationships between meteorological variables and turbine power generation, providing a reliable forecasting approach for operational wind energy management and supporting more stable integration of wind power into modern electricity systems.

The problem of depression among college students has become a burning research problem, as the number of psychosocial stress factors, academic loads, and lifestyle disorders that lead to the worsening of mental health has increased. Driven by the increasing need for innovative, data-driven, and interpretable diagnostic models, this paper presents a combined Machine Learning (ML) and metaheuristic optimization model for predicting depression using multidimensional psychosocial and academic data collected from 100 Computer Science students. The suggested hybrid model combines a Random Forest (RF) classifier with the Waterwheel Plant Algorithm (WWPA), a nature-inspired mechanism for optimizing hyperparameter settings and feature selection. Experimentation using the Random Forest baseline model yielded a baseline accuracy of 0.9081, a Sensitivity (True Positive Rate) of 0.8936, and an F-Score of 0.9032. The hybrid WWPA+Random Forest model showed significant gains after introducing the WWPA optimization method, achieving a high accuracy of 0.9577, a sensitivity of 0.9502, a specificity of 0.9644, and an F-score of 0.9553. These findings validate the high-quality performance of the proposed model in achieving balanced, high-precision classification and in resisting overfitting. The results highlight the potential to integrate ensemble learning with bio-inspired optimization to advance depression prediction, providing a scalable, explainable, and ethically appropriate framework for predicting depression early in a person’s life. This work opens the way to creating a proactive digital mental health system that will enable educational organizations to identify at-risk students early and offer timely, individualized support for well-being and academic achievement.

Forecasting financial markets remains a persistent challenge due to the nonlinear, stochastic, and nonstationary nature of stock price dynamics. This study is motivated by the need to enhance the robustness and adaptability of traditional statistical forecasting models through intelligent optimization. We propose an advanced hybrid framework that integrates the AutoRegressive Integrated Moving Average (ARIMA) model with the Fitness Greylag Goose Optimization (FGGO) algorithm—a refined metaheuristic inspired by collective behavioral intelligence and adaptive search strategies. The primary contribution of this research lies in the methodological fusion of classical time series modeling with dynamic metaheuristic optimization to improve predictive accuracy, convergence stability, and resistance to local optima. Comparative experiments on the historical stock prices of PT Bank Central Asia Tbk (BBCA.JK) demonstrate a substantial performance uplift: the baseline ARIMA model achieved a Mean Squared Error (MSE) of 0.0333, whereas the FGGO-optimized ARIMA reduced the MSE dramatically to 0.0038, outperforming other optimization techniques such as the Genetic Algorithm (GA), Whale Optimization Algorithm (WOA), and Particle Swarm Optimization (PSO). These results confirm that FGGO significantly enhances ARIMA’s capacity to capture intricate temporal dependencies and volatile market structures. The implications of this study extend beyond finance, offering a scalable, explainable, and high performance optimization paradigm for diverse time series forecasting applications in economics, engineering, and intelligent decision-support systems.

The explosive growth of digital marketing data and the increasing need for accurate revenue forecasting have driven the adoption of advanced Machine Learning (ML) techniques capable of modeling complex, nonlinear relationships in dynamic environments. Motivated by the limitations of traditional linear forecasting methods, this study proposes an optimized predictive framework that integrates the Extreme Gradient Boosting (XGBoost) algorithm with a novel metaheuristic, Dipper Throated Optimization (DTO), to enhance model performance on temporal marketing data. The key contribution of this work lies in combining ensemble learning with bio-inspired optimization to achieve superior predictive accuracy and stability in Time-Series forecasting tasks. As the experiments of the Digital Marketing Metrics dataset demonstrate, the original XGBoost model achieved a Mean Squared Error (MSE) of 0.0905 and a coefficient of determination (R2) of 0.8007, and the optimized XGBoost+DTO model has significantly improved results, with an MSE of 0.0010 and a coefficient of determination (R2) of 0.9002. These results support the argument that DTO is effective in hyperparameter optimization and reducing generalization errors. The results of this paper are not unique to digital marketing, and the authors have presented a scalable, interpretable optimization model that can be generalized to other data intensive fields, such as financial analytics, demand forecasting, and customer behavior modelling. The study is a good step in the right direction of creating more accurate, adaptive and data-driven decision-making in the digital economy by integrating ML and nature-inspired optimization.

The fact that educational, demographic, and macroeconomic variables interact nonlinearly has remained a thorn in the flesh of socio-economic analytics to date, making it challenging to forecast unemployment with sufficient precision. To address this, the current study presents a hybrid metaheuristic, Dipper Throated Optimization with Stochastic Fractal Search (DTOSFS), coupled with the Category Boosting (CatBoost) algorithm to improve predictive modelling. The suggested DTOSFS-CatBoost system combines the general exploratory search of DTO with SFS refinement to stochastic local optimization of hyperparameters, and alleviates overfitting. Empirical experiments have shown that whereas the original CatBoost gave results with a Mean Squared Error (MSE) of 0.0256 and Root Mean Squared Error (RMSE) of 0.1601 with a correlation coefficient of 0.873, the CatBoost optimized by DTOSFS had drastically better results with an MSE of 0.00033, RMSE of 0.00207, and a correlation coefficient of 0.930. These results confirm an increased exploration-to-exploitation ratio in DTOSFS and yield small, powerful designs that substantially enhance model stability, precision, and convergence speed. These results show that educational attainment (at least tertiary and primary enrollment) and demographics (at least the birth rate) are influential factors in unemployment variation. This addition to predictive performance is not the only one, and it provides a predictive data-driven labor-market optimization paradigm that can be replicated and interpreted. The research observes that hybrid metaheuristics and gradient boosting can be used to drive next generation economic intelligence systems for adaptive policy formulation and to enhance online, privacy conscious, and cross-domain unemployment prediction.