Enhancing Data Integrity in Mobile Crowdsensing Environment with Machine Learning and Cost-Benefit Analysis

Abstract

Mobile Crowdsensing (MCS) is a major source of a vast dataset containing heterogeneous types of data collected from various sources and stored in the local or remote server. Proper analysis of MCS data helps in better decision-making. However, MCS data suffers from data integrity issues, such as validity, accuracy, and reliability, that affect decision-making. Therefore, ensuring data integrity in the MCS environment is essential as it is a major source of a huge dataset. The proposed work considers user review data collection and analysis using a mobile application developed for the purpose. To ensure the data integrity, identification of fake and invalid reviews in the dataset need to be determined. This work proposes two approaches to solve data integrity issues. The first approach is to detect and eliminate fake/ invalid reviews from the dataset. The second is to identify the sources of fake/ invalid reviews and block them to protect the dataset from future fake reviews. Machine learning (ML) models are proposed to solve these issues and to ensure data integrity by filtering out fake reviews from real-time data sets. The proposed model uses data fuzzification over a purely mathematical model that categorizes users or customers as honest, suspicious, or malicious and their reviews/ feedback as genuine or fake using ratings provided by the user in the MCS Environment. Using the developed mobile application, user can give feedback about the desired location through various devices, which is stored in a cloud platform. The dataset can be analyzed through a fuzzy logic-based mathematical model followed by an ML algorithm and cost-benefit analysis to detect genuine reviews for maintaining data integrity. Further accuracy of the proposed models is compared with popular ML algorithms such as Naive Bayes (NB), Bayes Net(BN), Support Vector Machine(SVM), Decision Tree(J48), and Random Forest(RF). Initially, it achieves 99.79% of accuracy using the Random Forest algorithm that has been enhanced to 100% using cost-benefit analysis in cross-validation mode.