The convolutional neural networks (CNN) have shown promising results for various classification problems over the past years. However, selecting various CNN architectures is still challenging as each architecture performs differently with the same dataset. This research aims to evaluate the dependence of brain MRI on various predictive models of CNN based on the complexity of the data for Brain Tumor and Alzheimer’s Disease. Our proposed approach has three parts. First part is the pre-processing of the data which mainly focuses on class balancing and the estimation of data complexity. The second part uses stratified k-fold cross-validation for more reliable results. The last part corresponds to the implementation of four CNN models applying described methods. This paper compares the classification performance of rigorous experimentation on four CNN variants namely S-CNN (CNN trained from scratch), ResNet50, InceptionV3, and Xception over two brain MRI image datasets evaluated with and without the use of Principal Component Analysis (PCA). The work benchmarks CNN models by comparing the average scores of Accuracy, Precision, Recall, F1 score, and AUC score from the stratified five-fold cross-validation.

An accurate amount of fertilizer according to the real-time context is the basis of precision agriculture in terms of sustainability and profitability. Many fertilizers recommendation systems are proposed without considering the real-time context in terms of soil fertility level, crop type, and soil type. The major obstacle in developing the real-time context-aware fertilizer recommendation system is related to the complexity associated with the real-time mapping of soil fertility. Furthermore, the existing methods of determining the real-time soil fertility levels for the recommendation of fertilizer are costly, time-consuming, and laborious. Therefore, to tackle this issue, we propose a machine learning-based fertilizer recommendation methodology according to the real-time soil fertility context captured through the Internet of Things (IoT) assisted soil fertility mapping to improve the accuracy of the fertilizer recommendation system. For real-time soil fertility mapping, an IoT architecture is also proposed to support context-aware fertilizer recommendations. The proposed solution is practically implemented in real crop fields to assess the accuracies of IoT-assisted fertility mapping. The accuracy of IoT-assisted fertility mapping is assessed by comparing the proposed solution with the standard soil chemical analysis method in terms of observing Nitrogen (N), Phosphorous (P), and Potassium (K). The results reveal that the observations by both methods are in line with a mean difference of 0.34, 0.36, and -0.13 for N, P, and K observations, respectively. The context-aware fertilizer recommendation is implemented with the Logistic Regression (LR), Support Vector Machine (SVM), Gaussian Naïve Bayes (GNB), and K-Nearest Neighbor (KNN) machine learning models to assess the performance of these machine learning models. The evaluation of the proposed solution reveals that the GNB model is more accurate as compared to the machine learning models evaluated, with accuracies of 96% and 94% from training and testing datasets, respectively.

The increasing popularity of mobile phones has led to an abundance of online reviews, making it challenging for consumers to make well-informed purchasing decisions. This study proposes a novel recommendation system-based mobile phone rating classification approach using federated learning and Term Frequency-Inverse Document Frequency (TF-IDF) features. We created a novel dataset by scraping over 13,000 mobile phone reviews from Flipkart’s website. The proposed approach involves the development of a federated deep neural network (FDNN) to classify the newly created Flipkart dataset. This approach includes data cleaning, balancing, TF-IDF feature extraction, and prediction using federated learning. We employ two clients and one server and conduct three rounds of experiments. The experimental results demonstrate that the proposed approach achieved an accuracy of 96.68% on the aggregated server side while maintaining the security of customer data on their local devices. The proposed approach has the potential to assist consumers in making well-informed purchasing decisions and can be extended to other e-commerce platforms with large datasets of online reviews.

Aerial image-based target object detection has several glitches such as low accuracy in multi-scale target detection locations, slow detection, missed targets, and misprediction of targets. To solve this problem, this paper proposes an improved You Only Look Once (YOLO) algorithm from the viewpoint of model efficiency using target box dimension clustering, classification of the pre-trained network, multi-scale detection training, and changing the screening rules of the candidate box. This modified approach has the potential to be better adapted to the positioning task. The aerial image of the unmanned aerial vehicle (UAV) can be positioned to the target area in real-time, and the projection relation can convert the latitude and longitude of the UAV. The results proved to be more effective; notably, the average accuracy of the detection network in the aerial image of the target area detection tasks increased to 79.5%. The aerial images containing the target area are considered to experiment with the flight simulation to verify its network positioning accuracy rate and were found to be greater than 84%. This proposed model can be effectively used for real-time target detection for multi-scale targets with reduced misprediction rate due to its superior accuracy.

A brain tumor is a lethal neurological disease that affects the average performance of the brain and can be fatal. In India, around 15 million cases are diagnosed yearly. To mitigate the seriousness of the tumor it is essential to diagnose at the beginning. Notwithstanding, the manual evaluation process utilizing Magnetic Resonance Imaging (MRI) causes a few worries, remarkably inefficient and inaccurate brain tumor diagnoses. Similarly, the examination process of brain tumors is intricate as they display high unbalance in nature like shape, size, appearance, and location. Therefore, a precise and expeditious prognosis of brain tumors is essential for implementing the of an implicit treatment. Several computer models adapted to diagnose the tumor, but the accuracy of the model needs to be tested. Considering all the above mentioned things, this work aims to identify the best classification system by considering the prediction accuracy out of AlexNet, ResNet 50, and Inception V3. Data augmentation is performed on the database and fed into the three convolutions neural network (CNN) models. A comparison line is drawn between the three models based on accuracy and performance. An accuracy of 96.2% is obtained for AlexNet with augmentation and performed better than ResNet 50 and Inception V3 for the 120th epoch. With the suggested model with higher accuracy, it is highly reliable if brain tumors are diagnosed with available datasets.

This study proposed an AlexNet-based crowd anomaly detection model in the video (image frames). The proposed model was comprised of four convolution layers (CLs) and three Fully Connected layers (FC). The Rectified Linear Unit (ReLU) was used as an activation function, and weights were adjusted through the backpropagation process. The first two CLs are followed by max-pool layer and batch normalization. The CLs produced features that are utilized to detect the anomaly in the image frame. The proposed model was evaluated using two parameters—Area Under the Curve (AUC) using Receiver Operator Characteristic (ROC) curve and overall accuracy. Three benchmark datasets comprised of numerous video frames with various abnormal and normal actions were used to evaluate the performance. Experimental results revealed that the proposed model outperformed other baseline studies on all three datasets and achieved 98% AUC using the ROC curve. Moreover, the proposed model achieved 95.6%, 98%, and 97% AUC on the CUHK Avenue, UCSD Ped-1, and UCSD Ped-2 datasets, respectively.