Showing 5 results for Deep Learning
Mohammad Reza Mehregan, Arman Rezasoltani, Amir Mohammad Khani, Ali Hosseinzade Kashan,
Volume 0, Issue 0 (10-2025)
Abstract
In the modern industrial view, it is strategically imperative to predict failure of industrial machinery with a view to reducing the occurrence of unexpected failures and enhancing operational efficiency. This study seeks to introduce a new hybrid machine learning model for predictive maintenance, combining the use of deep learning and advanced ensemble machine learning models. The model presented follows a stacking ensemble structure, where XGBoost, CatBoost, Gradient Boosting, and a deep neural network are base learners. Thereafter, the LightGBM, acting as a meta-model, is used to collect its predictions. Further, in this study, the Optuna hyperparameter optimization framework is employed to optimize the hyperparameters automatically, and the NearMiss algorithm solves the class imbalance problem by enhancing the representation of the minority class and removing the bias in favor of the majority class. As can be seen in the experimental results, the combined model outperforms the single models, achieving an outstanding accuracy of 96.17%. This is followed by a precision of 97.86%, a recall of 94.4%, and an F1 score of 96.1%. It is worth noting that though the XGBoost models' independent results were high (with an F1 score of 89/41) and better than the 16 individual models studied in this paper and regarded as a comparison to the hybrid model, the hybrid model significantly defeated the independent models by nearly 7 percentage points, hence the strong suit of the smart ensemble framework in model combination. The model has been tried using industrial data with 10000 records of a milling machine system, which is representative of most industrial machinery. The model aids in making decisions in preventive maintenance processes in a more informed and timely way by detecting failures accurately before they happen, avoiding unwanted situations of unplanned downtime and operation costs. One can arrive at the conclusion based on these results that the mentioned hybrid model can offer a solid and workable way of predicting failures in the industrial context and can also be integrated into the actual maintenance processes without any issues.
Sangapu Venkata Appaji, R Shiva Shankar, K.v.s. Murthy, Chinta Someswara Rao,
Volume 31, Issue 3 (9-2020)
Abstract
Cancer is a consortium of diseases which comprises abnormal increase in cells growth by having potential to occupy and attack the entire body. According to study breast cancer is the most likely occurs in the women and which became the second biggest cause of women death. Due to its wide spread and importance some of the researchers work on this, but still there is a need to improvement. During this work in order to partially fulfill this proposed technique of deep learning along with RNN in predicting breast cancer disease which will help the doctor while diagnosis the patient. To assess the efficiency of the proposed method we used breast cancer data belong to UC Irvine repository. Precision, recall, accuracy and f1 score of proposed method shows good scores and proposed technique performs well Consortium
Amirhossein Masoumi, Rouzbeh Ghousi, Ahmad Makui,
Volume 33, Issue 3 (9-2022)
Abstract
Purpose: Non-cancerous prostate lesions such as prostate calcification, prostate enlargement, and prostate inflammation cause too many problems for men’s health. This research proposes a novel approach, a combination of image processing techniques and deep learning methods for classification and segmentation of the prostate in CT-scan images by considering the experienced physicians’ reports.
Methodology: Due to the various symptoms and nature of these lesions, a three-phases innovative approach has been implemented. In the first phase, using Mask R-CNN, in the second phase, considering the age of each patient and comparison with the standard size of the prostate gland, and finally, using the morphology features, the presence of three common non-cancerous lesions in the prostate gland has investigated.
Findings: A hierarchical multitask approach is introduced and the final amount of classification, localization, and segmentation loss is 1%, 1%, and 7%, respectively. Eventually, the overall loss ratio of the model is about 9%.
Originality: In this study, a medical assistant approach is introduced to increase diagnosis process accuracy and reduce error using a real dataset of abdominal and pelvics’ CT scans and the physicians’ reports for each image. A multi-tasks convolutional neural network; also presented to perform localization, classification, and segmentation of the prostate gland in CT scans at the same time.
Theodore Alvin Hartanto, Seng Hansun,
Volume 35, Issue 3 (9-2024)
Abstract
One method to diagnose retinal diseases is by using the Optical Coherence Tomography (OCT) scans. Annually, it is estimated that around 30 million OCT scans are performed worldwide. However, the process of analyzing and diagnosing OCT scan results by an ophthalmologist requires a long time so machine learning, especially deep learning, can be utilized to shorten the diagnosis process and speed up the treatment process. In this study, several pre-trained deep learning models are compared, including EfficientNet-B0, ResNet-50V2, Inception-V3, and DenseNet-169. These models will be fine-tuned and trained with a dataset containing OCT scanned images to classify four retinal conditions, namely Choroidal Neovascularization (CNV), Diabetic Macular Edema (DME), Drusen, and Normal. The models that have been trained are then tested to classify the test set and the results are evaluated using a confusion matrix in terms of accuracy, recall, precision, and F1-score. The results show that the model with the best classification results in the batch size of 32 scenario is the ResNet-50V2 model with an accuracy value of 98.24%, precision of 98.25%, recall of 98.24%, and F1-score of 98.24%. While for the batch size of 64, the EfficientNet-B0 model is the model with the best classification results with an accuracy value of 96.59%, precision of 96.84%, recall of 96.59%, and F1-score of 96.59%.
Arifa Khan, Saravanan P,
Volume 35, Issue 3 (9-2024)
Abstract
Optimizing production in the plastic extrusion industry is a pivotal task for small scale industries. To enhance the efficiency in today’s competitive market being a small-scale manufacturer over their peers is challenging. With the limited resources, having constraints on manpower, capital, space, often facing fluctuations in demand and production, simultaneously maintaining high quality became very important for the success. Among the plethora of KPIS used in manufacturing, Overall Equipment Effectiveness (OEE) stands out as corner stone. In this study, we collected real-world data from a plastic extrusion company. i.e., an HDPE Pipe manufacturing company. It serves as the backdrop for our study, this is based on the plastic extrusion sector and set out a goal of enhancing OEE through a comparative investigation of various ML models. To forecast and estimate OEE values, we used various Machine Learning models and examine each algorithm’s performance using metrics like Mean Squared Error (MSE) and model comparisons using the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), creating a comprehensive picture of each algorithm’s strength which enables the small businesses to make informed decisions and empowers them to stay agile and adapt to the changes in the manufacturing environment.
