Automated document classification through machine learning is crucial for handling extensive modern datasets, as manual analysis becomes impractical due to time and cost constraints. Real-world datasets often grapple with issues like class imbalance and high dimensionality, challenging classification algorithms. Our multi-class classification approach encompasses pre-processing, data balancing, feature engineering, and dimensionality reduction. We conducted experiments on seven datasets, including benchmarks and private data, leveraging natural language processing (NLP) techniques like lemmatization and stemming for corpus preparation. To mitigate feature dimensionality, we applied principal component analysis (PCA) and employed oversampling to tackle data imbalance. Finally, we utilized multilayer perceptron neural networks (MLP), support vector machines (SVM), and random forest (RF) models for clas-sification. Our framework achieved high accuracy, ranging from 77 % to 99 %, contingent on the dataset, while also demonstrating strong performance across additional metrics like precision, F1-score, and recall.

Drivers' behavior in traffic is a determining factor for the rate of accidents on roads and highways.This paper presents the design of an intelligent IoT system capable of inferring and warning about road traffic risks and danger zones, based on data obtained from the vehicles and their drivers mobile phones, thus helping to avoid accidents and seeking to preserve the lives of the passengers.The proposed approach is to collect vehicle telemetry data and mobile phone sensors data through an IoT network and then to analyze the driver's behavior while driving, along with data from the environment.The results of the inference serve to alert drivers about incidents in their trajectory as well as to provide feedback on how they are driving.The proposal is validated using a developed prototype to test its data collection and inference features in a small scale experiment.