FastForest: Increasing random forest processing speed while maintaining accuracy

Abstract Random Forest remains one of Data Mining’s most enduring ensemble algorithms, achieving well-documented levels accuracy and processing speed, as well regularly appearing in new research. However, with data mining now reaching the domain hardware-constrained devices such smartphones Internet Things (IoT) devices, there is continued need for further research into algorithm efficiency to deliver greater speed without sacrificing accuracy. Our proposed FastForest achieves this result through a combination three optimising components - Subsample Aggregating (‘Subbagging’), Logarithmic Split-Point Sampling Dynamic Restricted Subspacing. Empirical testing shows delivers an average 24% increase model-training compared whilst maintaining (and frequently exceeding) classification over tests involving 45 datasets on both PC smartphone platforms. Further show favourable results against number classifiers including implementations Bagging Subspace. With growing interest machine-learning mobile provides efficient classifier that can achieve faster smartphones.