Highly Accurate Hybrid Method for Attention Deficit Hyperactivity Disorder Classification Based on ANFIS-RFE-GWO

Interdisciplinary Description of Complex Systems 23(6), 632-657, 2025 HIGHLY ACCURATE HYBRID METHOD FOR ATTENTION DEFICIT HYPERACTIVITY DISORDER CLASSIFICATION BASED ON ANFIS-RFE-GWO Deepika*, Shaveta Arora and Meghna Sharma The NorthCap University, Department of Computer Science and Engineering Gurugram, Haryana, India DOI: 10.7906/indecs.23.6.3 Regular article Received: 10 July 2024. Accepted: 26 November 2025. ABSTRACT Attention Deficit Hyperactivity Disorder (ADHD) is one of the most prevalent neurobehavioral disorders, characterised by persistent patterns of inattention, impulsivity, and restlessness. This disorder significantly affects the personal, social, and academic development of individuals, with millions of children and adolescents worldwide experiencing its symptoms. Despite its high prevalence, accurate diagnosis is still a significant challenge for medical professionals, as distinguishing affected individuals from healthy controls is often complex. Many machine learning and deep learning approaches have been proposed for its diagnosis; the accuracy of ADHD diagnosis is still insufficient and needs further improvement. This study proposes a highly accurate hybrid framework to address this gap. The proposed method integrates the Recursive Feature Elimination technique to select the relevant and most significant feature subset, an Adaptive Neuro-Fuzzy Inference System to perform classification while handling inherent data uncertainty, and Grey Wolf Optimisation for hyperparameter tuning. Cross-validation is further applied to ensure optimal feature subset selection and robust model performance. The proposed framework is evaluated using phenotypic data from the ADHD-200 dataset, which included 547 patients diagnosed with the disorder and 325 healthy controls. For performance benchmarking, the proposed model is compared with several conventional machine learning classifiers, including Random Forest Classifier, K-Nearest Neighbour, and Gradient Boosting Classifier etc. Experimental results demonstrate that the proposed model outperforms several state-of-the-art approaches, achieving a classification accuracy of 98,30%, sensitivity of 96,82%, and F1-score of 97,60%. These results highlight the model’s potential as a reliable and effective diagnostic tool for clinical decision support, performing accurate detection and reducing misdiagnosis. KEY WORDS attention deficit hyperactivity disorder, adaptive neuro-fuzzy inference system, fuzzy logic, grey wolf optimisation, recursive feature elimination CLASSIFICATION JEL: I12 PACS: 87.57.R*Corresponding author, : ; +91 9582895328; *157, Sec-14, Gurugram,122001, Haryana, India Highly accurate hybrid method for attention deficit hyperactivity disorder classification ... INTRODUCTION Attention Deficit Hyperactivity Disorder (ADHD) is a prominent neuro-behavioural ailment of the modern age. It affects 5-9% of kids and adolescents worldwide [1] and may persist in adulthood as well. ADHD is the third most common mental disorder after depression and anxiety. Various studies indicate a significant increase in ADHD global prevalence from 1997 to 2022. An alarmingly high prevalence rate of 9,4% was observed in US kids in 2016 [2], and approximately 366,33 million adult ADHD cases were reported worldwide in 2020 [3]. The prevalence of ADHD among young children further increased to 10,47% in 2022 [4]. ADHD has frequent comorbidity with several other mental disorders, like schizophrenia, anxiety, depression, and sleep disorders [5-7]. Lack of concentration, increased activity level, and impulsiveness are the most noticeable symptoms of ADHD. Based on the symptoms, three different categories of ADHD have been identified, namely: i) Predominantly inattentive, ii) Predominantly hyperactive, and iii) ADHD-combined. ADHD imposes significant adverse effects on individuals. Children affected by ADHD are more likely to engage in disruptive conduct at school, face problems in making friends and relationships, and fulfil official commitments. They also have poor academic performance, resulting in a higher dropout rate in graduation and higher studies. Such children easily get engaged in increased usage of drugs, suicide attempts, and several other life-impairing activities, leading to a high risk of increased mortality rate. ADHD places a substantial financial burden of an estimated $122,8 billion on society [8] due to unemployment, productivity loss, and healthcare service requirements. RESEARCH MOTIVATION At present, there is no individual lab test or objective diagnostic method to diagnose ADHD. The diagnosis process mainly relies on the behavioural symptoms [9] observed by parents and teachers. Many questionnaire-based rating scales are used to assess the IQ level of the patient. The collected data is then analysed by medical experts to interpret the results. This approach is quite subjective as there is no golden standard for the assessment of results. Medical experts render diagnoses based on their experience, knowledge, and personal interpretation. Different medical experts may provide different diagnoses, often leading to misdiagnoses/overdiagnoses of ADHD. Another significant challenge in diagnosing ADHD arises due to the uncertainty linked to its symptoms. ADHD is heterogeneous; different patients can exhibit different symptoms altogether, posing complexities for accurate diagnosis by healthcare professionals. The presence of overlapping symptoms further compounds the diagnostic uncertainty. Various ADHD symptoms overlap with several other psychological disorders (e.g., schizophrenia, anxiety, depression, etc.), blurring the lines between symptoms and complicating the diagnostic process. Hence, there is a pressing need for an automated diagnostic process that is objective, precise, and reliable for ADHD, given its global prevalence and severity of impairments. Many studies [10-12] have been proposed for ADHD classification based on patients’ clinical data, yet the desired level of accuracy remains to be achieved. This is primarily due to the discrete approach adopted in most of the existing studies for ADHD classification. These studies have not factored in the inherent uncertainty in the data. Owing to heterogeneity and overlapping of its symptoms, it is not possible to quantify ADHD data accurately, resulting in substantial uncertainty and ambiguity within the dataset. This study seeks to effectively tackle these uncertainties and put forward a highly precise and efficient model for ADHD classification. MAJOR CONTRIBUTIONS This study proposes a highly accurate, reliable, and efficient hybrid model based on metaheuristic and fuzzy logic for ADHD classification. It utilises the personal characteristic 633 Deepika, S. Arora and M. Sharma data present in the benchmark ADHD-200 dataset [13]. Medical datasets generally contain many redundant or insignificant features that could create noise in the data and substantiall (...truncated)