English proficiency assessment in AR/VR

The ImmersiveLangPro dataset comprises comprehensive learner interaction records collected from immersive English language training platforms that utilize Augmented and Virtual Reality (AR/VR) technologies. It is specifically structured to support the development and evaluation of privacy-preserving, explainable AI models for English proficiency classification across diverse demographic and regional learner populations.

This dataset captures multimodal learning behavior across 252,782 user sessions, spanning sensor data, interaction metrics, and language skill assessments. Learners originate from varied educational backgrounds and geographical regions, and use different AR/VR-capable devices. The dataset is particularly suited for federated learning scenarios, as it enables client-side training simulations by incorporating client-level distributions across key features.

🧩 Feature OverviewThe dataset is organized into the following major categories:

Learner & Device Metadatauser_id, age_group, gender, native_language, education_level, country_region

device_type, internet_stability_score

Session & Interaction Metricssession_duration, number_of_sessions, avg_interaction_depth, environment_type

gesture_count, response_latency, task_completion_rate, replay_count, user_feedback_score

AR/VR Sensor Featureseye_gaze_duration, head_movement_variance, attention_span_score

navigation_accuracy, voice_command_usage

Language Skill Assessment Featuresreading_accuracy, listening_comprehension_score, speaking_fluency_score

pronunciation_score, vocabulary_usage_score, writing_score

grammar_accuracy, translation_skill_score

Target Labels for Learning ModelsCEFR_level: Official CEFR levels (A1–C2) as categorical labels

proficiency_score: Continuous proficiency scores (0–100) for regression tasks

classified_proficiency: Simplified classification (Low, Intermediate, High)

Auxiliary task-specific outputs: task_prediction_speaking, task_prediction_listening, task_prediction_writing

This dataset is ideal for benchmarking federated, explainable, and multimodal AI models aimed at advancing personalized English language education in immersive digital ecosystems.