Random forest method for estimation of brake specific fuel consumption

Abstract

The internal combustion engine is a widely used power equipment in various fields, and its energy utilization is measured using brake specific fuel consumption (BSFC). BSFC map plays a crucial role in the analysis, optimization, and assessment of internal combustion engines. However, due to cost constraints, some values on the BSFC map are estimated using techniques like K-nearest neighbor, inverse distance weighted interpolation, and multi-layer perceptron, which are recognized for their limited accuracy, particularly when dealing with distributed sampled data. To address this, an improved random forest method is proposed for the estimation of BSFC. Polynomial features are employed to increase higher dimensions of features for random forest by nonlinear transformation, and critical parameters are optimized by particle swarm optimization algorithms. The performance of different methods was compared on two datasets to estimate 20%, 30%, and 40% of BSFC data, and the results reveal that the method proposed in this paper outperforms other common methods and is suitable for estimating the BSFC map.

Comparing machine learning techniques for software requirements risk prediction

“…The model's confusion matrix revealed accurate predictions across different risk levels: four (4) data points correctly identified as Catastrophic, eleven (11) as High, eighteen (18) as Moderate, thirty-three (33) as Low, and seven (7) as Insignificant. The model achieved an accuracy rate of 98%, with a 95% confidence interval and a Kappa value of 97%.…”

“…It is an ensemble ML method [18] that allows for regression and classification. It consists of a set of DTs constructed using bagging, whose variation is controlled based on the training data.…”