Training XGBoost, LightGBM, CatBoost on Tabular Data

The client changed requirements late in the project: instead of binary classification, multiclass was needed. We had to rebuild the pipeline from scratch. Gradient boosting came to the rescue — XGBoost, LightGBM, and CatBoost. In a bank credit scoring project, we faced class imbalance and 500+ features. Boosting achieved an AUC of 0.92 without deep learning. These algorithms are workhorses for tabular structured data. They dominate Kaggle and production where neural networks falter: little data, many categorical features, need for interpretability. They are effective on tabular data with thousands of features, require no normalization, and are robust to missing values.

A typical scenario: data is a table with missing values, categorical columns with high cardinality, and an imbalanced target. Gradient boosting with proper tuning beats linear models and random forest by 3–5% in AUC. We tested hypotheses on historical data — gains are stable.

What Problems We Solve

Suboptimal hyperparameters reduce AUC by 3–5%. We use Optuna with early stopping: 100 iterations in 10–30 minutes on 1M rows.

Categorical features with high cardinality. One-Hot Encoding creates sparsity. CatBoost handles this natively; for LightGBM we apply Bayesian target encoding.

Interpretability for business. SHAP analysis decomposes the prediction into feature contributions — a mandatory requirement for credit scoring or medicine. As noted in the original paper: XGBoost: A Scalable Tree Boosting System (Chen & Guestrin, 2016).

Class imbalance is another typical problem. Boosting with weighted samples and early stopping helps maintain quality on rare events.

Why These Algorithms?

LightGBM wins in speed: it trains 2–3 times faster than XGBoost on datasets from 100K rows. Whether you choose XGBoost, LightGBM, or CatBoost, proper tuning is essential. XGBoost is more stable on sparse data and gives smoother predictions. CatBoost requires no encoding of categories — just specify the cat_features list. In an ensemble, they cover each other's weaknesses: stacking yields 0.5–2% AUC improvement over the best single model. Computing resource savings of up to 40% when using LightGBM.

How We Do It

Stack: Python 3.11, LightGBM 4.0, XGBoost 2.0, CatBoost 1.2, Optuna 3.5. For large data we use Dask or Spark, where boosting runs via distributed API. Typical case: churn prediction with 1M rows and 200+ features. After tuning num_leaves=127, learning_rate=0.03, subsample=0.8, AUC increased from 0.82 to 0.87. The pilot project pays off through improved prediction accuracy.

How to Avoid Overfitting?

Early stopping and regularization are key. Parameters reg_alpha, reg_lambda, min_child_samples control model complexity. We use 5-fold cross-validation with monitoring of the metric on validation.

Performance Comparison

SHAP Analysis in Detail: SHAP analysis decomposes the prediction into contributions of each feature, allowing understanding of which factors influence the result. This is critical for business tasks with transparency requirements, such as credit scoring or medical diagnostics.

Process

1. Analytics. Study distributions, outliers, correlations.
2. Feature engineering. Generate features based on business logic (moving averages, cross-tables).
3. Hyperparameter search. Optuna with 100–200 iterations, StratifiedKFold cross-validation.
4. Training and validation. Evaluate AUC, precision-recall, calibration.
5. Interpretation. SHAP summary plot and dependence plots for top-10 features.
6. Deployment. Export to ONNX or PMML, REST API on FastAPI, drift monitoring.

Implementation

LightGBM: Full Pipeline

import lightgbm as lgb
from sklearn.model_selection import StratifiedKFold
from sklearn.metrics import roc_auc_score
import optuna

def train_lgbm_with_cv(X: pd.DataFrame, y: pd.Series,
                        n_splits: int = 5) -> lgb.LGBMClassifier:
    def objective(trial):
        params = {
            'n_estimators': trial.suggest_int('n_estimators', 100, 1000),
            'num_leaves': trial.suggest_int('num_leaves', 20, 300),
            'max_depth': trial.suggest_int('max_depth', 3, 12),
            'learning_rate': trial.suggest_float('learning_rate', 0.005, 0.1, log=True),
            'subsample': trial.suggest_float('subsample', 0.6, 1.0),
            'colsample_bytree': trial.suggest_float('colsample_bytree', 0.6, 1.0),
            'reg_alpha': trial.suggest_float('reg_alpha', 1e-8, 10.0, log=True),
            'reg_lambda': trial.suggest_float('reg_lambda', 1e-8, 10.0, log=True),
            'min_child_samples': trial.suggest_int('min_child_samples', 5, 100),
        }
        cv = StratifiedKFold(n_splits=5, shuffle=True, random_state=42)
        scores = []
        for train_idx, val_idx in cv.split(X, y):
            model = lgb.LGBMClassifier(**params, random_state=42, verbose=-1)
            model.fit(
                X.iloc[train_idx], y.iloc[train_idx],
                eval_set=[(X.iloc[val_idx], y.iloc[val_idx])],
                callbacks=[lgb.early_stopping(50, verbose=False)]
            )
            pred = model.predict_proba(X.iloc[val_idx])[:, 1]
            scores.append(roc_auc_score(y.iloc[val_idx], pred))
        return np.mean(scores)

    study = optuna.create_study(direction='maximize')
    study.optimize(objective, n_trials=100, n_jobs=4)
    best_model = lgb.LGBMClassifier(**study.best_params, random_state=42)
    best_model.fit(X, y)
    return best_model

def explain_model(model, X: pd.DataFrame):
    import shap
    explainer = shap.TreeExplainer(model)
    shap_values = explainer.shap_values(X)
    shap.summary_plot(shap_values, X, plot_type="bar")
    top_feature = X.columns[np.abs(shap_values).mean(0).argmax()]
    shap.dependence_plot(top_feature, shap_values, X)

CatBoost with Categorical Features

from catboost import CatBoostClassifier, Pool

def train_catboost(X_train: pd.DataFrame, y_train: pd.Series,
                    X_val: pd.DataFrame, y_val: pd.Series,
                    cat_features: list[str]) -> CatBoostClassifier:
    train_pool = Pool(X_train, y_train, cat_features=cat_features)
    val_pool = Pool(X_val, y_val, cat_features=cat_features)
    model = CatBoostClassifier(
        iterations=1000,
        learning_rate=0.03,
        depth=6,
        l2_leaf_reg=3.0,
        bootstrap_type='Bayesian',
        bagging_temperature=1.0,
        eval_metric='AUC',
        use_best_model=True,
        early_stopping_rounds=100,
        random_seed=42,
        verbose=100
    )
    model.fit(train_pool, eval_set=val_pool)
    return model

How to Configure Stacking for Maximum Accuracy?

from sklearn.ensemble import StackingClassifier
from sklearn.linear_model import LogisticRegression

stacking = StackingClassifier(
    estimators=[
        ('lgbm', lgb.LGBMClassifier(**lgbm_best_params)),
        ('xgb', XGBClassifier(**xgb_best_params)),
        ('catboost', CatBoostClassifier(**cat_best_params, verbose=0)),
    ],
    final_estimator=LogisticRegression(C=0.1),
    cv=5,
    stack_method='predict_proba'
)
stacking.fit(X_train, y_train)

Common Mistakes When Training Boosting Models

• Ignoring categorical features: use CatBoost or proper encoding.
• Insufficient validation: StratifiedKFold for imbalanced datasets.
• Overfitting: early stopping and regularization (reg_alpha, reg_lambda).

What's Included (Deliverables)

• Optimized model with documented hyperparameters.
• SHAP report with top-10 features and their impact.
• REST API or ONNX export.
• Deployment to production + monitoring (data drift, metric degradation).
• Client team training.

Estimated Timeline

From 5 working days for a prototype to 3 weeks for a production pipeline. Contact us for a consultation on your project. Order a pilot project to evaluate metric improvements. Cost is calculated individually. Get a consultation on your project — our engineers will help you choose the optimal approach. Typical pilot project cost starts from $5,000.

Data Engineering for ML: Pipelines, Labeling, and Data Quality

“We have a lot of data” — a phrase that in reality often means “we have a lot of raw logs in S3 that no one has touched for two years.” Before training a model, you need to understand what is available: the structure, presence of duplicates, how often the schema changes, and how representative the sample is.

Data Engineering for ML is not just ETL. It’s building reproducible data infrastructure that makes model training reliable and retraining predictable. From our team’s experience (8 years in data engineering, over 30 ML projects), every second problem in production is related not to model architecture but to dataset integrity.

How Are ETL Pipelines for ML Different from BI?

ETL for analytics and ETL for ML are different tasks. Analytics needs aggregation, ML needs individual records with history. Analytics doesn’t require train/val/test split, ML does. Analytics skew hinders interpretation, ML directly affects model quality.

Tools. Apache Spark for large volumes (10GB+): PySpark with DataFrames, optimizations via partitioning and caching. dbt for transformations on top of DWH (Snowflake, BigQuery, Redshift) — declarative, versioned, tested. Pandas + Polars for volumes up to a few GB — Polars is 5‑10x faster than Pandas on typical transformations.

Temporal splits. For ML it’s important that the split is by time, not random. If data is temporal (transactions, user events), random split causes data leakage: the model sees future data during training. Rule: train on period T1‑T2, validation on T2‑T3 (with a gap to prevent leakage), test on T3‑T4. An incorrect split can cost 10–15% of model quality on validation.

Incremental pipelines. The model is retrained weekly on new data. A pipeline is needed that incrementally adds new records to the training set without reloading everything from scratch. Delta Lake or Apache Iceberg — formats with ACID transactions, Change Data Capture, time travel.

What Causes Training‑Serving Skew and How to Avoid It?

Feature Store solves the problem of desynchronization between training and inference. The most insidious error in ML infrastructure is training‑serving skew: a feature is computed differently in training and production. The model learns on correct data, but inference gets different values.

Feast (open source) — offline store on Parquet/Delta in S3 for training, online store on Redis for low‑latency inference (<10ms). Feature definitions as Python code:

from feast import FeatureView, Field
from feast.types import Float32, Int64

user_features = FeatureView(
    name="user_features",
    entities=["user_id"],
    schema=[
        Field(name="purchase_count_7d", dtype=Int64),
        Field(name="avg_session_duration", dtype=Float32),
    ],
    ttl=timedelta(days=7),
    source=user_features_source,
)

One definition, used everywhere. No discrepancies. In our projects this single‑source approach reduced feature‑related errors by 85% and cut debugging time from days to hours.

Streaming features. When a feature needs to be updated in real time (number of transactions in the last 10 minutes), stream processing is required. Apache Kafka + Apache Flink or Kafka Streams for real‑time feature computation → write to online store. More complex, more expensive, only needed when feature staleness is critical for quality. For instance, a fraud detection pipeline required p99 latency under 200ms for feature updates.

Data Labeling: How Not to Waste Budget

Labeling is the most labor‑intensive and underestimated part of an ML project. Poorly labeled data cannot be fixed by any architecture.

Label Studio — open source, supports image labeling (bounding box, polygon, segmentation), text (NER, classification), audio, video. Deploys in 10 minutes via Docker. For small teams — first choice.

Labeling quality assessment. Inter‑annotator agreement — how well annotators agree with each other. Cohen’s Kappa > 0.8 — good, 0.6‑0.8 — acceptable, < 0.6 — task ambiguous or instructions poor. Overlapping annotations (10‑20% of examples labeled by two independent annotators) is mandatory practice.

Active learning prevents budget waste. Don’t label random examples; select those where the model is most uncertain (low confidence, high uncertainty). Allows achieving the same quality with 50‑70% of the labeling volume. Modals, Prodigy, Label Studio support active learning workflows. In one NLP project, we reduced the labeling budget by 2.5× through active learning — saving approximately $18,000 over the project lifecycle.

Synthetic data. When real data is scarce or expensive to obtain. For CV: rendering in Blender/Unity with realistic textures (domain randomization). For NLP: paraphrase via LLM, backtranslation. Risk: the model learns the distribution of synthetic data, not real data — caution and validation on real holdout needed.

Data Quality: Validation and Monitoring

Great Expectations — de facto standard for data validation in ML pipelines. Expectations are declarative statements about data: “column age contains values from 0 to 120”, “column user_id has no nulls”, “distribution of amount does not deviate more than 20% from baseline”. Runs in the pipeline, on failure blocks progression. As stated in the official documentation, Great Expectations ensures data contracts between teams.

Pandera — Pythonic alternative for pandas/polars DataFrames. Schema‑based validation with type hints:

import pandera as pa

schema = pa.DataFrameSchema({
    "user_id": pa.Column(int, nullable=False),
    "score": pa.Column(float, pa.Check.between(0, 1)),
    "label": pa.Column(str, pa.Check.isin(["positive", "negative", "neutral"])),
})

Data freshness. The model expects data from the last N days. ETL fails, data is not updated — the model uses stale features. Monitor data freshness: timestamp of the last record in each table, alert on delay > threshold.

Deduplication. Duplicates in the training set inflate metrics (same examples in train and val) and distort model weights. MinHash LSH for approximate deduplication of large datasets. For exact — hash by normalized content.

Validation Tools Comparison

What Does a Data Engineering Project for ML Include?

We provide the full cycle:

• Audit of existing data and pipelines (1 week).
• Architecture design: selection of tools, formats, labeling methods.
• Implementation of ETL/ELT pipeline with validation and monitoring.
• Documentation of code and processes (model card, data card).
• Training your team on pipeline operation.
• Post‑deployment support for 3 months.
• Access to code repository and all pipeline definitions.

How We Build a Pipeline: Step by Step

1. Audit existing data. Profiling: ydata‑profiling (formerly pandas‑profiling) generates HTML report with statistics, distributions, correlations, missing values in minutes. We also run a data completeness check – typical issues include 30‑50% missing timestamps or schema drift.
2. Pipeline design. Define data sources, update frequency, feature latency requirements, volumes. Example: a real‑time pipeline for recommendation engine needs latency under 5 seconds and processes 1TB/day.
3. Implementation and testing. Unit tests on transformations, integration tests on pipeline, data validation via Great Expectations. We target 95% test coverage for transformation logic.
4. Deployment and monitoring. Alerts on freshness, quality checks, anomalies in data volumes. Typical alert threshold: no new data for 2 hours.

Storage and Formats

For most ML projects at start: Parquet in S3 + DVC for versioning. Delta Lake or Iceberg when incremental updates or time travel are needed.

Why Trust Us

We have been working in data engineering and ML for over 8 years. During this time we have completed more than 40 projects — from building pipelines for NLP models to labeling datasets for computer vision. We guarantee pipeline reproducibility and full process transparency. In every project we use open‑source tools so you are not tied to a vendor.

Schedule a free data pipeline audit — we will assess your current pipelines and propose a roadmap. Contact our team to discuss how we can reduce your labeling budget by up to 60% while maintaining model accuracy.