ShoroBanglaAir-IMU

This dataset contains IMU sensor recordings of all 11 Bengali vowel characters written in the air by 250 volunteers. Data were captured using a finger-mounted NGIMU sensor at 100 Hz, recording 3-axis gyroscope and 3-axis accelerometer signals. Each participant wrote each vowel twice in a single session, yielding 5,511 raw CSV samples organized into 11 class folders. Each CSV file contains approximately 500 rows of 7-column motion data (timestamp, gyroscope X/Y/Z, accelerometer X/Y/Z). Magnetometer and barometric channels available on the device were excluded to retain only motion-relevant signals.Participants were 250 native Bengali-speaking volunteers aged 20–35 (mean ≈ 26 years). Baseline experiments using Random Forest and other shallow classifiers achieved 60–70% accuracy across 11 classes, confirming meaningful discriminative signal in the data. The dataset is intended for research in air-writing recognition, gesture-based input, Bengali script recognition, and time-series classification benchmarking.