• Poorvaa Wasnik
• Nakul Indurkar
• Ayush Sarode
• Pranay Kailuke
• Preeti Dudhe

Automated bot accounts, networks engaged in coordinated inauthentic behaviour (CIB), and AI-synthesised profile photographs collectively represent a growing threat to the credibility of online information ecosystems, financial markets, and democratic processes. Contemporary detection solutions operate in isolation, lack transparent decision rationales, and frequently augment their outputs with fabricated taxonomic labels that misrepresent ground truth. This paper introduces ForensIQ, a full-stack, open-architecture social media forensics platform built to close these gaps simultaneously. ForensIQ integrates a 26-feature extraction pipeline with a three-model machine learning ensemble (Random Forest, XGBoost, and LightGBM) blended against a 22-rule heuristic scorer to classify Twitter/X profiles as Real, Suspicious, or Fake. A principled NaN-encoding strategy preserves the informational difference between absent and genuinely zero engagement, enabling adaptive model weighting based on data completeness. Louvain community detection on a semantically grounded similarity graph exposes CIB clusters, while a multi-quality Error Level Analysis (ELA) module authenticates profile images without relying on EXIF metadata stripped by hosting platforms. SHAP TreeExplainer provides per-feature attribution for every classification. Trained and validated on the publicly available Cresci-2017 benchmark (7,642 profiles, 80/20 split), LightGBM achieves 99.41% accuracy and 99.99% ROC-AUC. The platform is delivered as a React 18 single-page application over a FastAPI/MongoDB backend, with automated PDF report generation. All classifications are grounded exclusively in verifiable signals; the system enforces an honest-emptiness policy that prohibits fabricated or demo-mode outputs.

Bot Detection, Social Media Forensics, Fake Account Classification, Coordinated Inauthentic Behaviour, Deepfake Detection, Ensemble Learning, SHAP Explainability, XGBoost, LightGBM, Random Forest, Louvain Community Detection, Sentence Embeddings, Cresci-2017

"ForensIQ: A Multi-Modal Social Media Forensics Platform", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.13, Issue 10, page no.a476-a482, October-2026, Available :http://www.jetir.org/papers/JETIR2610051.pdf

"ForensIQ: A Multi-Modal Social Media Forensics Platform", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.13, Issue 10, page no. ppa476-a482, October-2026, Available at : http://www.jetir.org/papers/JETIR2610051.pdf