.png){kind=icon}
.png){kind=icon}
Return
A Novel Ensemble Model for Optimizing Author Profiling
.png){kind=icon}
OA
AI
DOI：10.1109/ojcs.2026.3687421.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Author Profiling (AP) is a specialized area of digital forensic science that influences linguistic stylistics, semantics, and syntax to identify an author's age, gender, occupation, and educational background. AP plays a vital role in various domains, including forensic linguistics and social media analysis, especially by law enforcement organizations to find suspect or anonymous writers. Additionally, it uses linguistic pattern analysis to assist academic integrity, targeted marketing, and the detection of fraudulent accounts. AP plays a vital role in various domains, including marketing, forensics, security, and medicine. Although prior research has covered English, Arabic, and French extensively, Roman Urdu lacks representation in the literature. This research uses the Fire’18 Mapon SMS dataset to predict the age and gender of authors from their Roman Urdu text messages. A variety of linguistic features were employed, including sentence-based, word-based, and character-based elements, along with their combinations. The proposed system presents an ensemble model for AP, leveraging AdaBoostM1 and Random Forest algorithms, collectively referred to as ABMRF. Several model AdaBoostM1, J48, NB-Updatable, KNN, RF, CHIRP, and NB served as benchmarks for evaluating the ABMRF framework. Results indicate that ABMRF outperforms these models across multiple feature sets. Using combination of features highest results achieved 56.28% for age while for gender RF outperform achieving 74.571%. Using character-based features, the proposed system achieved an accuracy of 54% for age, while 74.77% for gender. For word-based features, the accuracy was 50.57% for age, while 60% for gender. Sentence-based features yielded 49.71% accuracy for age prediction and 62% for gender prediction. Notably, CHIRP and KNN models demonstrated the poorest performance across all linguistic feature sets for predicting age and gender.
Keywords:
Fire’18 MaponSMS
AdaBoostM1
random forest
ensemble learning
and author profiling for security
optimizing author profiling
digital forensic
Journal
I
IF:
8.2
Papers:
411
Citations:
810
View more organizations