Semi-Supervised Classifier for Ride-Hailing Malicious Trip Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Ride-hailing platforms face challenges in accurately detecting malicious activities due to insufficient training data, as the number of malicious trips is small and often not comprehensive, leading to ineffective safety checks and information security breaches.
Innovation Solution
A semi-supervised machine learning algorithm is employed to identify malicious trips by training a classifier using historical data, deploying it to classify new trips, and continuously re-training based on newly detected data samples, with a staging database for data cleansing and confidence-score-based Time-To-Live assignments to filter false positives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If offline learning methods are used to train machine learning models based on historical data, then the model can detect malicious trips, but the training data is insufficient because the number of malicious trips in historical data is small and not comprehensive
Solution Approach 1:
The system performs preliminary classification of trips as malicious or benign using an initially trained classifier, then uses these preliminary results to expand the training dataset. This preliminary action creates additional training samples that overcome the scarcity of actual malicious trip data in historical records.
Solution Approach 2:
A staging database is introduced as an intermediary component between the classifier and the training data. The staging database stores classified trips and enables the system to iteratively expand training data by converting classification results into new training samples, thereby bridging the gap between limited historical data and comprehensive training requirements.
2Productivity
If the classifier is deployed to classify new trips continuously, then new malicious trip samples can be obtained for re-training, but false positives may be incorrectly classified as malicious
Solution Approach 1:
Trips classified as malicious are first stored in a staging database before being used for re-training. This preliminary storage step allows for verification and validation, preventing hasty incorporation of potentially false positive samples into the training dataset.
Solution Approach 2:
The system implements a feedback mechanism where classified trips are stored in the staging database, reviewed, and only verified malicious trips are used to re-train the classifier. This feedback loop ensures that false positives are filtered out before affecting model updates, maintaining reliability while enabling continuous improvement.
Data Source
AI summary
Methods, systems, and apparatus for detecting malicious activities in a ride-hailing platforms are described. An exemplary method comprises: identifying a set of trips from historical data to form training data; training a classifier based on a plurality of features of the set of trips in the training data to identify whether a given trip is malicious or benign; deploying the classifier to classify new trips in the ride-hailing platform for a first period of time to obtain a plurality of malicious trip candidates; storing the plurality of malicious trip candidates in a staging database for a second period of time for data cleansing based on supplementary data collected during the second period of time; fetching, from the staging database, a set of malicious trip candidates that have been stored in the staging database longer than the second period of time; and re-training the classifier.