Performance evaluation of machine learning based voting classifier system for human activity recognition

Abstract

In the last few decades, Human Activity Recognition (HAR) has been a centre of attraction in many research domains, and it is referred to as the potential of interpreting human body gestures through sensors and ascertaining the activity of a human being. The present work has proposed the voting classifier system for human activity recognition. For the voting classifier system, five machine learning classifiers are considered: Logistic Regression (LR), K-Nearest Neighbour (KNN), Random Forest (RF), Naive Bayes (NB), and Support Vector Machine (SVM). These machine learning classifiers are ensembled by analyzing the best performers among them. The ensemble voting classifiers are proposed under two variations, i.e., hard voting and soft voting. The various combinations of voting classifiers are compared and evaluated. For experiments, the benchmark dataset of the UCI-HAR dataset is considered, and all the data files are combined into a single file to avoid bias. The dimensionality of the dataset is reduced by using Principal Component Analysis (PCA) from 561 features to 200 components. The results reveal that Voting Classifier-II (a combination of SVM, KNN, and LR) using soft voting outperformed other machine learning classifiers.

Predicting Frequent-Infrequent Access of Objects in Tiered Cloud Storage Services

“…In our assessment, we scrutinize nine distinct machine learning models, alongside the online algorithm introduced in [1]. For brevity, each method is denoted by an acronym: Online Algorithm (ONL), Random Forest (RF) [18], Linear Regression (LR) [19], k-Nearest Neighbors (KNN) [20], Decision Tree (DCT) [21], Support Vector Machine with RBF (SVM-R) [22], and Sigmoid (SVM-S) kernels [23], and committee algorithms such as Soft Voting Standard (SV) [24] and Hard Voting (HV) [25]. We harness the Sklearn library for implementing the machine learning models, adopting standard parameters.…”

Adaptive Analysis of Machine Learning Algorithms on Medical Datasets for Disease Prediction

“…Therefore, two neighboring intervals constitute different classes. This algorithm has two main phases: the training phase and the classification phase [17].…”

Real-time human activity recognition from smart phone using linear support vector machines

“…We also showed the method of classification adopted in the treatment. [25] SVM-Gaussian kerne 96.50% De Leonardis et al [26] K-nearest neighbor -feedforward neural network -SVM -decision tree -Naïve Bayes 97.00% Nurhanim et al [27] SVM polynomial kernel -one versus all 98.57% Agarwal and Alam [28] SVMs-k-nearest neighbor-linear discriminant analysis 98.00% Minarno et al [29] SVM+LR 98.00% Jindal et al [30] SVM, KNN, and LR 92.78% Patel and Shah [31] Long short-term, LR 92.00% Navita and Mittal [32] SVM 98.03% Figure 6. Model accuracy scores…”