TY  - JOUR
AU  - Paneru, Biplov 
AU  - Paneru, Bishwash 
AU  - Shah, Krishna Bikram 
AU  - Shrestha, Awan 
AU  - Poudyal, Ramhari 
AU  - Poudyal, Khem Narayan 
PY  - 2024
TI  - Autism Spectrum Disorder Prediction Using Hybrid Deep Learning Model and a Recommendation System Application for Autistic Patient
JF  - Journal of Computer Science
VL  - 20
IS  - 9
DO  - 10.3844/jcssp.2024.1040.1050
UR  - https://thescipub.com/abstract/jcssp.2024.1040.1050
AB  - The goal of this study is to create machine learning models that use a big dataset to predict Autism Spectrum Disorder (ASD). To achieve optimal performance, a number of algorithms were employed and refined, including Support Vector Machines (SVM), Random Forest, XGBoost, Multi-Layer Perceptron (MLP) and a hybrid model that combines MLP and SVM. To evaluate the durability of the model's performance, the study used cross-validation and hyperparameter tuning techniques. Measures such as memory, accuracy, precision and F1-score have been employed to assess how well the models predict ASD. It's interesting to note that the RBF kernel did quite well in the grid search using the SVM model. All models produced good findings, with test set accuracies ranging from 87-97%. With 97% accuracy on the testing set, the CatBoost algorithm demonstrated excellent performance. Additionally, the hybrid MLP + SVM model demonstrated the potential benefits of combining different approaches by doing well on both the training and testing sets. Additionally, a Flask application was made to provide a straightforward user interface for the machine learning models that were learned. For those with ASD or who are at risk, this application generates predictions based on user input and provides tailored recommendations and interventions. The work highlights the potential for developing useful tools to support ASD diagnosis and intervention, as well as the effectiveness of machine learning techniques in ASD prediction. The robustness and applicability of the existing models may be strengthened by more research and validation on bigger and more varied datasets