An Improved Ensemble Model Using Random Forest Branch Clustering Optimisation Approach


  • Ayinla, I. B.
  • Akinola, S. O.


Decision Tree Forest, Random Forest, Classification and Prediction


The world of technology is growing faster and helping organisations to repositioning their focus and vision for business. The introduction of Internet of Things (IoTs) devices has contributed in no small measure to business values and the world livelihood. The need for efficient Machine Learning Algorithms (MLAs) to drive these devices to perform to optimal or near optimal has been a serious challenge. The inadequacies of these MLAs has resulted in loss of trust and sometimes led to legal litigation against Artificial Intelligent (AI) organisations.
Hence, we introduced a novel approach to improving traditional Random Forest RF, an ensemble model, which is known to be high performance classifier using branch clustering Random Forest (BCRF) technique in Decision Tree Forests (DTFs). The sensitivity, specificity and F-score values as well as extra pruning of pessimistic after Entropy and Information Gain Ratio (IGR) were used to isolate the weaker groups for model improvement. The model produced more accurate results with a better speed of execution when used on the same dataset as Naïve Bayes, RandomForest and K-nearest Neighbour.