LightGBM

The purpose of the present study is to use the ensemble learning model to combine the predictions of Random Forest (RF), Long-Short Term Memory (LSTM), and Recurrent Neural Network (RNN) models for the Top and Bottom Prices of Bitcoin. To this aim, in the first stage, Bitcoin's top and bottom prices are predicted using three machine learning models. In the second stage, the outputs of the models are presented as feature variables to the Extreme Gradient Boosting (Xgboost) and Light Gradient Boosting Machine (LightGBM) models to predict the price tops and bottoms. Then, in the third stage, the outputs of the second stage are combined through the voting ensemble classifier pattern to predict the next top and bottom prices. The data of top and bottom Bitcoin prices in the 1-hour time frame from 1/1/2018 to the end of 6/30/2022 are used as target variables and 31 technical analysis indicators as feature variables for the three models in the first stage. 70% of the data is regarded as learning data, and the remaining 30% is considered for the second and third stages. In the second phase, 50% of the data is considered for learning the output of the previous stage and 50% for the test data. Finally, the prediction values are evaluated with real data for the three models and the proposed ensemble learning model. The results reveal the improvement of the performance, precision, and accuracy of the ensemble model compared to weak learning models.

Relying on the Area Under the Curve (AUC) measure, we compare the performance of the Logit regression model and the LightGBM algorithm. Despite these methods being common in the literature, our study emphasizes the role of statistical inference to evaluate and compare the results comprehensively. We use the training set of the Vesta (2018) dataset, provided by Vesta—a global fraud prevention company headquartered in the United States specializing in payment solutions and risk management. Originally released as part of a Kaggle competition focused on credit card fraud detection, this dataset comprises diverse transaction records, representing a rich source for exploring advanced fraud detection methods. Our analysis reveals that while the LightGBM algorithm generally yields higher predictive accuracy, the differences between the calculated AUCs of the two methods are not statistically significant. This underscores the importance of using inferential techniques to validate model performance differences in fraud detection.

In today’s data-driven hospitality sector, customer interactions increasingly occur through digital platforms, generating extensive behavioral and transactional information. This study analyse the prediction of Customer Lifetime Value (CLV) using machine learning models—Linear Regression, Random Forest, and LightGBM—trained on features derived from hotel website interactions and booking records. After comprehensive data preprocessing, the models were evaluated using MAE, RMSE, and R² metrics. LightGBM achieved the highest predictive performance (R² = 0.504), followed by Random Forest (R² = 0.497), while Linear Regression underperformed (R² = 0.386), highlighting the advantages of non-linear models in modeling intricate customer patterns. Residual analyses confirmed LightGBM's stability and low bias across diverse customer profiles. Apart from prediction, the study applies Recency-Frequency-Monetary (RFM) analysis to segment customers into distinct value-based groups. These segments form the basis for tailored marketing strategies, allowing hotels to allocate resources more efficiently, enhance customer retention, and develop targeted campaigns aligned with customer potential. By integrating web-derived behavioral data with advanced modeling and segmentation, this research offers hotel managers practical tools for strategic planning in customer relationship management.