Obstructive sleep disordered breathing (SDB) is classically defined as a syndrome of “upper airway dysfunction during sleep, characterised by snoring and/or increased respiratory effort secondary to increased upper airway resistance and pharyngeal collapsibility. Pediatric SDB may cause substantial morbidity that affects multiple target organs and systems [1]. Treatment relies on a multidisciplinary approach involving craniofacial surgeons and pulmonologists when surgery fails to improve SDB. In particular some children may need ventilatory support with nocturnal ventilators. Since treatment efficacy is mandatory in this setting, many patients undergo routine sleep exams, to titrate pressures and settings. This approach is time-consuming and demand frequent hospitalisations. However, using built-in software data of these ventilators allow to have a breath by breath analysis of respiratory flows and pressures. Our work seeks to create an automatic analysis of these signals extracted from the ventilators, and provide clinicians a simple and ready to use tool to control treatment efficacy. To solve this problem, we have thus proposed a hybrid approach, that couples a Long Short-Term Memory (LSTM) Neural Network with an ensemble of Machine Learning models. The advantage of this novel approach lies in combining the strengths of both methods. LSTM networks are particularly effective in capturing the temporal dynamics and sequential nature of respiratory signals [2]. On the other hand, machine learning models are particularly effective at handling handcrafted or statistical features, enabling them to capture global patterns and distinctive signal characteristics [3]. We have then trained and tested our method on a benchmark dataset, composed of overnight sleep recordings of pediatric patients suffering from SDB. Our results indicate the effectiveness of the proposed methodology in automatic extracting apnea events with an accuracy of 97% and an area under receiver operator characteristic curve of 98%. This hybrid approach also outperforms the results obtained using only one method at a time (LSTM or the ML ensemble), demonstrating the strength and synergy achieved through their integration. The presented approach represents thus a novel deep learning framework for automated detection of SDB in children using respiratory signals.