This paper explores the use of attention-focused eye gaze analysis to predict autistic traits using transfer learning. The study focuses on two main phases: the first phase involves implementing various transfer learning algorithms to predict ASD traits using open-source image datasets from Kaggle and Zenodo, with fivefold cross-validation to ensure reliability. The ConvNextBase model is found to have the best performance, achieving 80.4% accuracy on Kaggle and 80.71% on Zenodo. The second phase introduces a novel method to measure attentiveness using the HOG and Linear SVM-based face detectors to calculate the Eye Aspect Ratio (EAR) and determine if participants are attentive. The system plots EAR values over time to assess attentiveness, with a threshold of 0.20 for 100 consecutive frames indicating un-attentiveness. The paper discusses the methodology, experimental setup, and results, highlighting the potential of using attention as a novel feature in diagnosing autism.This paper explores the use of attention-focused eye gaze analysis to predict autistic traits using transfer learning. The study focuses on two main phases: the first phase involves implementing various transfer learning algorithms to predict ASD traits using open-source image datasets from Kaggle and Zenodo, with fivefold cross-validation to ensure reliability. The ConvNextBase model is found to have the best performance, achieving 80.4% accuracy on Kaggle and 80.71% on Zenodo. The second phase introduces a novel method to measure attentiveness using the HOG and Linear SVM-based face detectors to calculate the Eye Aspect Ratio (EAR) and determine if participants are attentive. The system plots EAR values over time to assess attentiveness, with a threshold of 0.20 for 100 consecutive frames indicating un-attentiveness. The paper discusses the methodology, experimental setup, and results, highlighting the potential of using attention as a novel feature in diagnosing autism.