Hybrid FCNN-LSTM Approach for Autism Spectrum Disorder Classification

Abstract

Several symptoms can be observed to detect a complicated Neuro Developmental Disorder (NDD) named Autism Spectrum Disorder (ASD). The social interactions, behaviour, and communication were greatly impacted by this disease. Early Detection (ED) of the disease will facilitate in effective treatment, so this ED is crucial. Here, the conventional methods face difficulty, because it depends on expert evaluations. These expert evaluations are subjective and time-consuming. These conventional ASD diagnostic methods are manual, subjective and suspectible to inconsistencies. The development of reliable treatment ways is further complicated, because these methods are slow and it may generate unreliable diagnosis. The complexity and variability of the ASD symptoms are not effectively captured by the current Machine Learning (ML) methods. This will result in inaccurate predictions. For ASD classification, a Hybrid Fuzzy CNN-LSTM approach was suggested for resolving those limitations. Data preprocessing is the initial process in this method, as it includes Noise Reduction (NR) and normalization. Thus high-quality inputs were also ensured. Then, the spatial patterns in data can be detected by employing Convolutional Neural Networks (CNN), and it can be used for performing Feature Extraction (FE). Here, these features are then fed into an Long Short-Term Memory (LSTM) network for detecting temporal relations. Also, Fuzzy Logic (FL) was also employed in this study, and it facilitate in managing the ASD data’s uncertainty and unpredictability. At last, the accuracy and accessibility was improved. The spatial analysis, temporal analysis, large data management of FL are collaborated, and these collaboration facilitates the hybrid model in enhancing the classification performance. Diabetes datasets usually include patient data with diagnostic characteristics linked to ASD, such as social interaction, behavioural, and communication metrics. Following metrics like Accuracy (Acc) , Precision (P), and Recall (R) were utilized for assessing the suggested model. From the outcomes of the simulation, it is clear that the suggested Hybrid Fuzzy CNN-LSTM model performs better than the current ML methods in terms of performance metrics. Thus, an improved diagnostic accuracy of ASD was attained.