Machine Learning-Based Classification of ASD for Improved Diagnostic Accuracy

Autism Spectrum Disorder(ASD) is a neuro-developmental disorder that influences how people interact with others and communicate hence affecting their ability to interact with others. This is an indication that with the use of primary symptoms it is quite difficult for humans to predict/estimate the present condition and stage of ASD. The main goal is to achieve early diagnosis of ASD in a person using different machine learning and deep learning models. In fact, for humans, it becomes challenging to compiled such amount of data as well as use it to predict via various results. But with the help of machine learning, this task becomes more efficient and less time consuming. This paper aims to show several measurements related to ASD using various classifiers such as Support Vector Machines(SVM), Random Forest Classifier (RFS) and Decision trees. The main purpose to use Random forest algorithm is because it contains the object called feature importance which calculates the importance of the various parameters used in the model between 0(low) to 1(high). Also, random forest helps in prediction using multiple decision trees by selecting the output with high probability. SVM algorithm is used because it is efficent in handling dataset of high dimension i.e. Dataset having large number of features. SVM is also considered best algorithm in terms of memory managament. As one of the main problems in making machine learning model is overfitting, SVM serves it the best in case of handling and reducing overfitting of the dataset. Decision trees are used because the main advantage of decision tree is that it goes through several inputs, make multiple predictions based on the result of each decision tree and gives the best output that follows. There are various advantages of using decision trees such as ease of understanding and easy to interpret and understand by others. Deep learning is also used by using layers like Sequential, Dense and Dropout which goes deep into data to find required solutions and insights. The proposed model achieves an accuracy of 86.25% using random forest algorithm, 84.38% using Support Vector Machines algorithm and 81.25% using decision trees algorithm. Out of the three, it is clear that Random forest algorithm works well out of the three algorithms.