Dynamic gesture recognition with data reduction and machine learning algorithms using Soli dataset

As technology continues to advance, the expectations of technology users are also evolving. Smart electronic devices have significantly altered many daily habits of individuals, making them increasingly dependent on sensors. Among these sensors, radars have gained widespread use in daily life due to their advantages such as operating in a noncontact manner, being unaffected by lighting and weather conditions, and not infringing on privacy. In 2016, Google introduced the Soli project. Soli project has presented a dynamic hand gesture recognition system with a radar of very small dimensions. In this study, the Soli dataset-comprising 11 distinct hand gestures and 5,225 samples-has been utilized. Classical techniques such as truncation and downsampling have been applied to reduce data dimensions, and the classification has been performed using various machine learning algorithms, including Random Forest, k-Nearest Neighbors, Radial Basis Function Support Vector Machine, Decision Tree, and Gradient Boosting. Although deep learning algorithms often yield high performance with large datasets, in this study, the Random Forest algorithm achieved the highest classification accuracy with $\mathbf{8 3. 2 7 \%}$.