Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder with social interaction, communication, and repetitive behavior problems. Early diagnosis is the key to facilitating on-time intervention and enhancing patient outcomes. This paper is an evaluation of comparative performance of different machine learning (ML) and deep learning (DL) algorithms for early ASD diagnosis based on behavioural and demographic information from AQ-10 screening tests. Supervised machine learning algorithms like K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Decision Tree, Random Forest, XGBoost, and Logistic Regression are compared with unsupervised cluster algorithms like K-Means, Agglomerative, Spectral, Gaussian Mixture Models (GMM), and BIRCH. Deep learning models consisting of Dense layers, Dropout, Batch Normalization, Long Short-Term Memory (LSTM), Residual connections, and Sigmoid activation functions are investigated. Random Forest Classifier is used for feature selection, which asserts that behavioural markers are the most powerful predictors for ASD, and demographic indicators are not very influential. The comparative analysis indicates the strengths and weaknesses of ML and DL algorithms, and the best performing model is chosen, and that model is used to develop an Autism Care Hub platform for early screening, diagnosis, and continuous monitoring of ASD. Experimental results show the efficacy of ensemble ML techniques and DL networks and uncover their potential in real-world clinical deployment.