Filtered Crowdsourced Starlink Latency and Throughput Measurements with Satellite Density and Weather Features

This dataset contains crowdsourced Starlink satellite internet performance measurements collected from two open measurement platforms, MLab NDT7 and Cloudflare AIM, spanning January to November 2025 across multiple countries. The raw measurements were filtered using two anomaly detection methods (percentile-based filtering for latency and Isolation Forest for throughput) and enriched with additional features, including Starlink satellite visibility density computed from publicly available TLE orbital data, client-server geographic distance, and hourly weather conditions (temperature, precipitation, cloud cover, and wind speed) obtained from the Open-Meteo API. Each record includes the test timestamp, client and server locations, network performance metrics (download latency, throughput, jitter, and packet loss), along with all derived features. The dataset comprises approximately 11.6 million latency measurements and 11.9 million throughput measurements, organized as monthly CSV files in two directories corresponding to the two filtering methods. This data supports research on predicting Starlink network performance using machine learning models that leverage geographic, temporal, orbital, and meteorological features.