AI Driving Classification

Machine learning system that classifies driving behavior from smartphone sensor data.

Overview

A research project that combines Convolutional Neural Networks with Long Short-Term Memory networks into a ConvLSTM architecture to classify normal and aggressive driving behavior. The system ingests accelerometer, gyroscope and GPS data collected from smartphones, then applies normalization, noise filtering and maneuver segmentation. The ConvLSTM model reached 97.71% accuracy, outperforming Stacked LSTM and BiLSTM alternatives. Published in MDPI Future Transportation (2025) as 'Driving Behavior Classification Using a ConvLSTM'.

Role

Researcher & ML Engineer

Focus

  • MACHINE LEARNING
  • CONV LSTM
  • IoT

Talk to João about a project →