time-series-split

Temporal partitioning and backtesting for time-correlated datasets

Predicts the posted rate of a truckload from lane, distance, equipment, weight and pickup date. Chronological holdout, model comparison and a price range per prediction. Built for an ML engineer assessment.

Practical guide to validating time-series models: stationary vs non-stationary setups, leakage pitfalls, expanding/rolling TimeSeriesSplit, Ridge/LinearRegression baselines, and bootstrap CIs for metrics. Includes utilities for feature engineering (percent_change) and visualization.

Predicting House Prices — End-to-End Regression Workflow

Predicting NBA MVP vote totals across 42 seasons (17,976 player-seasons) with stacked and from-scratch adaptive ensembles — including why the hand-built weighting scheme lost to a plain average

Predict NYC restaurant inspection grades (A vs B/C+) with leak-safe, time-split ML baselines.

Offline Python research pipeline for network intrusion detection: audits original vs corrected CICIDS2017, uses chronological splits and validation-frozen random-forest thresholds, and reports results with their limitations. No live traffic.

Machine learning early warning for hunger crises in Nigeria's LGAs. Random forest, time-based split, Streamlit app.

Modelo que predice si un contrato publico peruano terminara siendo modificado, con features construidas sin fuga temporal y particion temporal estricta

以 2,069 萬筆電商行為與 109 萬則商品評論，完成漏斗診斷、意圖預測與損益兩平決策的可重現資料產品。

Cost-aware credit card fraud detection pipeline: time-based split, probability calibration, and business-aligned threshold tuning (AUPRC-first).