A Warwick University team introduced the RAVEN pipeline, which uses synthetic training data and a combined GBDT‑GP model to rank and validate TESS candidates, achieving over 97% AUC on all false‑positive scenarios, 91% overall accuracy on 1,361 external TOIs, and confirming 118 new exoplanets.
This article introduces a privacy-preserving gradient boosting decision tree (GBDT) solution built on multi‑party computation, detailing its background, training steps, the MPC tools used, and the Squirrel framework’s workflow, while discussing performance challenges and experimental results demonstrating scalability to millions of samples.
This article walks through using Apache Spark for large‑scale GBDT training, covering the challenges of massive data, Spark deployment, PySpark code examples, differences from Pandas, feature engineering, mmlspark installation, early‑stopping tricks, performance bottlenecks, and a systematic evaluation of alternative frameworks.
Time series forecasting, essential across weather, finance, and commerce, involves tasks like classification, clustering, anomaly detection, and especially prediction; this article explores its definitions, evaluation metrics, traditional methods, machine‑learning approaches, deep‑learning models such as TFT, and emerging AutoML tools, offering practical insights and best practices.
This article explains how a five‑line pandas script can generate frequency‑based encodings for categorical columns, why such encodings help low‑frequency categories, and how they integrate with feature crossing and industry use cases like ad exposure, PV/UV counting, and fraud detection.
This article analyzes why gradient‑boosted decision trees frequently outperform neural networks in many Kaggle contests, examining data characteristics, model strengths and weaknesses, real competition examples, and practical guidelines for choosing the right model based on nonlinearity and interpretability.
XGBoost Serving is an open‑source, high‑performance inference system built on TensorFlow Serving that adds dedicated servables for pure GBDT, GBDT+FM binary‑classification, and GBDT+FM multi‑classification models, providing automatic version lifecycle management, GRPC/HTTP APIs, and up to 50 % latency reduction, now available on GitHub after successful deployment in iQIYI’s recommendation platform.
This article explains the architecture and ranking process of modern recommendation systems, covering the two-stage pipeline of candidate generation and ranking, the evolution from rule‑based methods to logistic regression, GBDT, wide‑and‑deep, and deep learning models, and discusses challenges such as feature non‑linearity, multi‑objective optimization, and the need for post‑ranking interventions.
This article reviews the development of Zhihu's search system, describing the transition from early GBDT ranking to deep neural networks, the introduction of multi‑objective and position‑bias‑aware learning‑to‑rank methods, context‑aware techniques, end‑to‑end training, personalization, and future research directions.
This article dissects a ByteDance interview algorithm that asks for optimal partitioning of a sequence, derives the underlying GBDT split‑gain formula, and shows how to reduce the naïve O(n²) computation to O(n) using the variance‑gain trick employed by LightGBM.
By extracting hour‑level activity logs and training supervised models—including CART, GBDT, and neural networks—on user tags, the team simulated short‑term metrics for new reward campaigns, enabling earlier prediction of next‑day retention and shortening experiment cycles despite delayed T+1 data.
This article describes how a GBDT‑based prediction model was built to forecast file release application parameters such as volume ratio, target audience, and gray level, covering data collection, feature engineering, model training, service deployment, and practical considerations for handling bad cases.
iQIYI introduces RSLIME, a model‑agnostic, sample‑level feature importance method for its three‑stage small‑video recommendation system, enabling interpretable analysis of a complex ranking module that combines DNN, GBDT, and FM, and demonstrating stable, AUC‑correlated insights for optimization and feature selection.
This article outlines the end‑to‑end design of automotive finance risk‑control processes, discusses key data integrity and customer segmentation considerations, and details machine‑learning modeling practices—including logistic regression, decision trees, GBDT, XGBoost, LightGBM and CatBoost—along with an automated platform to streamline model development and deployment.
The article shares practical strategies for processing massive advertising data in the Tencent algorithm competition, covering model selection between GBDT and neural networks, efficient data partitioning methods for low‑resource environments, and the importance of feature engineering to achieve top rankings.
This article summarizes the Ant Financial paper presented at DASFAA 2018 that proposes a universal local explanation method for Gradient Boosting Decision Tree models, detailing the problem definition, the PMML‑based algorithm for attributing feature contributions, experimental validation on fraud detection data, and the practical benefits for model transparency and improvement.
The author, a Sun Yat‑sen University undergraduate and repeat weekly champion, shares practical tips on handling large datasets, effective feature engineering, and combining GBDT with a custom deepFFM model to achieve top scores in the Tencent advertising algorithm competition.
Gradient Boosting Decision Tree (GBDT) is an ensemble learning method that combines additive and gradient boosting, detailed with its mathematical foundations, regression and classification algorithms, implementation using scikit‑learn, and a real‑world consumer‑finance fraud detection case achieving high AUC and KS metrics.
This article describes a tourism recommendation system that addresses data sparsity, seasonality, and geographic variations by using an offline‑online architecture, GBDT+LR CTR prediction, exponential decay scoring, and extensive feature engineering, achieving a 1.6% conversion‑rate increase and high accuracy and recall.
This article introduces ensemble learning as a meta‑algorithm that combines multiple base classifiers, explains the two main strategies—Boosting and Bagging—covers their bias‑variance trade‑offs, outlines essential steps, and provides concrete examples such as AdaBoost, Random Forest, and GBDT applied to user age prediction.
By training Gradient Boosted Decision Trees on offline delivery data, extracting leaf‑node indices as one‑hot features, and merging them with merchant, traffic, and weather information, the study reduces overall ETA MAE by 3.4%, raises N‑minute accuracy by 2.2 points, and achieves larger gains for high‑value orders, demonstrating that GBDT‑derived features markedly improve food‑delivery time predictions.
The iQIYI recommendation system combines a two‑stage pipeline of recall and ranking, evolving from logistic regression to a GBDT‑FM‑DNN ensemble, using online feature storage, extensive feature engineering, and configurable strategies to deliver personalized video suggestions while addressing feature drift and multi‑objective business goals.
This article details a data-driven solution for the 2016 O2O coupon redemption competition, describing dataset partitioning, extensive feature engineering across user, merchant, and coupon dimensions, handling leakage, and model fusion using XGBoost, GBDT, and RandomForest, achieving top AUC scores through weighted ensemble.
The article describes a data‑driven approach for identifying low‑value new users in a food‑delivery platform by labeling 7‑day repeat‑purchase behavior, extracting order, behavior, merchant and user features, and training a combined Gradient Boosted Decision Tree and Logistic Regression model to improve fraud detection and merchant penalty decisions.
