PARKINSON'S DISEASE AND AUTISM SPECTRUM DISORDER DIAGNOSIS USING SIGNIFICANT FEATURES AND ENSEMBLE DEEP LEARNING

More than 6 million people globally suffer from progressive Neurological Disorders (NDD) such as Parkinson's disease (PD). Furthermore, a variety of developmental impairment disorders that have an impact on a patient's capacity for social interaction and communication are collectively referred to as Autism Spectrum Disorder (ASD). Conventional methods for detecting PD and ASD, however, are frequently manual and need expertise. To solve this issue, a prior study presented an Ensemble Feature Selection (EFS) method based on the Fuzzy-based Beetle Swarm Optimization Algorithm (FBSOA) and Score-based Artificial Fish Swarm Algorithm (SAFSA). This technique chooses the most significant features from the database to increase the PD detection rate. First, Min-Max Normalization (MMN) was used to normalize the dataset's scale. Dimensionality reduction (DR) is achieved by the application of Binomial Cumulative Distribution Function-based Principal Component Analysis (BCDPCA). The optimal features for PD classification are then found via Ensemble-based FS (EFS). To classify PD, Ensemble Learning (EL) classification approaches such as Fuzzy K-Nearest Neighbor (FKNN) hybrid classifier, Kernel Support Vector Machines (KSVM), Fuzzy Convolutional (NN) Neural Network (FCNN), and RF are employed. However, dataset which are used for this work has missing values which will leads to the misclassification error. And also, MMN does not handle outliers very well. Resolving those challenges, this study proposed an improved model for PD and ASD detection. Then first missing values are imputed using Most Common value (MCI). And data normalization will be done by ZSN. DR will be performed based on BCDPCA. Significant features are selected from the dimension reduced data using ensemble of Enhanced Chicken Swarm Optimization (ECSO) and Improved Whale Optimization (IWO). Finally, PD and ASD are detected using ensemble of FCNN and LSTM network. The suggested schema obtained better performances for obtained values of precision, recall, and f measure in experimental outcomes.