-
An Artificial Neural Network project dedicated to Classify Autistic Patients from Non Autistic Patients (Predict the likelihood of a person having autism using survey and demographic variables.)
-
The project highlightsthe significance of Data Cleaning (Preprocessing) ,Data Exploration (Descriptive Statistics) , and employs professional visualizations to enhance data understanding.
| Feature | Description |
|---|---|
| index | The participant’s ID number |
| AX_Score | Score based on the Autism Spectrum Quotient (AQ) 10 item screening tool AQ-10 |
| age | Age of participant |
| gender | 'm' for Male and 'f' for Female |
| ethnicity | Ethnicities in text form ['White-European', 'Latino', '?', 'Others', 'Black', 'Asian','Middle Eastern ', 'Pasifika', 'South Asian', 'Hispanic','Turkish', 'others'] |
| jaundice | 'no' and 'yes' for Whether or not the participant was born with jaundice? |
| autism | 'no' and 'yes' for Whether or not anyone in the immediate family has been diagnosed with autism? |
| country_of_res | Countries in text format |
| used_app_before | 'no' and 'yes' for Whether the participant has used a screening app |
| result | Score from the AQ-10 screening tool |
| age_desc | Age as categorical ['18 and more'] |
| relation | Relation of person who completed the test ['Self', 'Parent', '?', 'Health care professional', 'Relative','Others'] |
| Class/ASD | Participant classification ['NO', 'YES'] |
- Source : Autism Screening on Adults
- Preprocessing
-
- Handling Missing Values
- Handling Outliers
- Handling Categorical Features
- ANN sequential model with input layers of 10 neurons and activation function = 'relu' ,one hidden layer of 8 neurons and activation function = 'relu' and output layer of 1 neuron with activation function = 'sigmoid'
- Optimizer used 'Adam'
- Loss Function used 'binary_crossentropy'
- Metrics used to evaluate the model performance 'accuracy'