Optimized LSTM Based Framework with Time Series Data to Detect Outliers

Abstract

Outlier detection is a technique used to remove any anomalies that are contributed to a database. Outliers are extreme values that deviate from the normal. Significant variations from the results of the other datasets might be a sign of experimental mistakes, measurement inconsistencies or innovation. This study compiled several kinds of skin disorders and brain tumors in addition to analyzing contextual data extracted from the MRI and ISIC datasets, which contain numerous temporal dimensions of data series. This paper introduces an outlier detection of disease-based methodology for identifying outliers, which employs a fine-tuned hyper parameter of the Long Short-Term Memory (LSTM) model. Furthermore, 34680 images were processed after the data augmentation technique. The findings indicate that the recommended methodology improves the outcome of the suggested model when it comes to identifying outlier images in the context of tumours and skin lesions. The recommended model is compared with LSTM (Max-pool), LSTM (Average-pool), and LSTM (Min-pool) and achieves an accuracy of 94.54%. It is expected that the proposed system would benefit the medical community by making patient medical data storage, analysis, and distribution possible. Additionally, this approach guarantees that the provision of healthcare is patient-centered and sensitive to their preferences.