Non-Invasive Autism Detection Using MLP Neural Network

Abstract:Early detection of Autism Spectrum Disorder (ASD) is crucial for timely intervention and support. In this study, we propose a non-invasive autism detection system based on physiological and behavioral data collected from a non-invasive dataset. The system employs machine learning techniques, specifically a Multi-Layer Perceptron (MLP) neural network, to classify individuals as autistic or non-autistic. Data preprocessing, scaling, and encoding ensure high-quality inputs for the model. The proposed approach achieves high classification accuracy while maintaining robustness and generalization, as validated through train-test evaluation, confusion matrix analysis, and ROC curve assessment. The results demonstrate that machine learning can effectively support early, non-invasive autism screening, offering a promising tool for clinicians and caregivers. Project Description This project implements a non-invasive autism detection system using a Multi-Layer Perceptron (MLP) neural network. It processes physiological and behavioral data from a CSV dataset (datos_autismo_no_invasivo.csv) to classify individuals as autistic or non-autistic. The system includes data preprocessing, scaling, model training, and evaluation with performance metrics. Features Data Preprocessing: Handles non-numeric columns, encodes categorical variables, and scales features for training. MLP Neural Network: Trains a multi-layer perceptron with two hidden layers (64 neurons each) using backpropagation. Training Monitoring: Tracks loss and accuracy for both training and testing sets across iterations. Evaluation Metrics: Confusion matrix ROC curve and AUC Classification report (precision, recall, F1-score) Correlation heatmap of features Requirements Python 3.7 or higher Libraries: pandas, numpy, matplotlib, seaborn, scikit-learn Install dependencies using pip: