Autism spectrum disorder（ASD）is an abstruse brain disorder in neuroscience research,which lead by challenges of social interactions,speech and nonverbal communication,and repetitive behaviors.Diagnosis of ASD is mostly based on behavioral analysis,which is time-consuming and depends on patient cooperation and examiner expertise.The assessable analysis for ASD is rationally challenging due to the limitations of publicly available datasets.The non-invasive whole-brain scans endure the aptitude to succor for diagnosing neuropsychiatric disorder diseases such as autism,dementia,and brain cancer.For diagnostic or prognostic tools,functional Magnetic Resonance Imaging（f MRI）exposed affirmation to the biomarkers in neuroimaging research because of f MRI pickup inherent connectivity between the brain and regions.There are profound studies in ASD with introducing machine learning or deep learning methods that have manifested advanced steps for ASD predictions based on f MRI data.Machine learning（ML）and deep learning（DL）are recently applied to diagnosing autistic brain images and changing the questionnaire troublesome policy to treat a patient.There are profound studies in ASD with introducing machine learning or deep learning methods that have manifested advanced steps for ASD predictions based on f MRI data.However,there are some limitations using ML approaches such as investigation with functional magnetic resonance imaging（f MRI）,depend on region-based analysis,and handling with the big dataset.Moreover,utmost antecedent models have an inadequacy in their capacity to manipulate performance metrics such as accuracy,precision,recall,and F1-score,ROC curve,and AUC score.In this thesis,we introduce ML and DL approaches to get state-of-the-art results to overcome all of these challenges.We used the support vector machine（SVM）as the most popular ML classifier and deep neural network（DNN）to classify ASD,respectively.We have done these works after investigation a vast amount of research papers in the field of neuroimaging of autism disorder.In the first study,we proposed a novel architecture based on the Bag-of-Features model for ASD classification.In the beginning,we preprocess the images to extract the speeded-up robust features（SURF）from the selected feature point locations.The Bag-of-Feature extraction procedures include feature concatenation,select the most vital feature,and convert to feature vector.After that,we employ K-Means clustering to create a word visual vocabulary from the SURF vector.Then,we encode the Bag-of-Features by adopting coding and quantization techniques to get each class’ s indexed database.We prefer the most tremendous machine learning methods SVM as ASD classifier.Finally,we independently evaluate our proposed architecture’s performance using three different datasets from different fields,including ABIDE f MRI preprocessed images and subject’s face images.In our experiments,weigh against other state-of-the-art methods that our ML classifiers with Bag-of-Feature extractors reinforce in medication and clinical purposes of ASD.However,after completing the first work,we noticed that the classification accuracy is not good enough to rely on ML.Besides,SVM is not performing well to handle large datasets like as ABIDE dataset.In the second study,to avoid the circumstance of the first work problems and getting motivated to start the second work,we proposed a benchmark model,an avant-garde“Dark ASDNet,” which has the competence to extract features from a lower level to a higher level and bring out promising results.In this work,we considered 3D f MRI data to predict binary classification between ASD and typical control（TC）.Firstly,we preprocessed the 3D f MRI data by adopting proper slice time correction and normalization.Then,we introduced a novel Dark ASDNet which is surpassed the benchmark accuracy for the classification of ASD.Our model’s outcomes unveil that our proposed method establishes state-of-the-art accuracy of94.70% to classify ASD vs.TC in ABIDE-I,NYU dataset.Finally,we contemplated our model by performing evaluation metrics including precision,recall,F1-score,ROC curve,and AUC score,and legitimize by distinguishing with recent literature descriptions to vindicate our outcomes.The proposed Dark ASDNet architecture provides a novel benchmark approach for ASD classification using f MRI processed data.