There are numerous algorithms and solutions for car or object detection as humanity is aiming towards the smart city solutions. Most solutions are based on counting, speed detection, traffic accidents and vehicle classification. The mentioned solutions are mostly based on high-quality videos, wide angles camera view, vehicles in motion, and are optimized for good visibility conditions intervals. A novelty of the proposed algorithm and solution is more accurate digital data extraction from video file sources generated by security cameras in Bosnia and Herzegovina from M18 roadway, but not limited only to that particular source. From the video file sources, data regarding number of vehicles, speed, traveling direction, and time intervals for the region of interest will be collected. Since finding contours approach is effective only on objects that are mobile, and because the application of this approach on traffic junctions did not yield desired results, a more specific approach of classification using a combination of Histogram of Oriented Gradients (HOG) and Support Vector Machines (Linear SVM) has shown to be more appropriate as the original source data can be used for training where the main benefit is the preservation of local second-order interactions, providing tolerance to local geometric misalignment and ability to work with small data samples. The features of the objects within a frame are extracted first by standardizing the feature variables and then computing the first order gradients of the frame. In the next stage, an encoding that remains robust to small changes while being sensitive to local frame content is produced. Finally, the HOG descriptors are generated and normalized again. In this way the channel histogram and spatial vector becomes the feature vector for the Linear SVM classifier. With the following parameters and setup system accuracy was around 85 to 95%. In the next phase, after cleaning protocols on collected data parameters, data will be used to research asphalt deformation effects.

Security is one of the most actual topics in the online world. Lists of security threats are constantly updated. One of those threats are phishing websites. In this work, we address the problem of phishing websites classification. Three classifiers were used: K-Nearest Neighbor, Decision Tree and Random Forest with the feature selection methods from Weka. Achieved accuracy was 100% and number of features was decreased to seven. Moreover, when we decreased the number of features, we decreased time to build models too. Time for Random Forest was decreased from the initial 2.88s and 3.05s for percentage split and 10-fold cross validation to 0.02s and 0.16s respectively.

Clinical mistreatment and mismanagement are big issues caused by detection of too many false negative patients. Therefore, lung cancer diagnostic inaccuracy and methods to surpass it in a minimally invasive way is often the subject of research, as it is case of this study. This study focuses on the use of machine learning algorithms as a noninvasive tool to differentiate malignant pleural effusions from benign effusions. It provides performance comparisons between Adaptive neuro-fuzzy inference system (ANFIS), Support vector machine (SVM), RUS Boosted Tree (RUSBoost) and K-Nearest-Neighbor (K-NN) techniques for lung cancer detection. The proposed algorithms were chosen based on the current state of the art in the field of pulmonary diagnostics. The novelty of this work is the application of machine learning models for classification of lung cancer based on expression of tumor markers obtained from serum and pleural fluids. The performance of all models is compared and validated on data samples of 168 patients. Three classification model, SVM, RUSBoost and K-NN performed equally well, whereas underperforming model was ANFIS.

This research presents the epileptic focus region localization during epileptic seizures by applying different signal processing and ensemble machine learning techniques in intracranial recordings of electroencephalogram (EEG). Multi-scale Principal Component Analysis (MSPCA) is used for denoising EEG signals and the autoregressive (AR) algorithm will extract useful features from the EEG signal. The performances of the ensemble machine learning methods are measured with accuracy, F-measure, and the area under the receiver operating characteristic (ROC) curve (AUC). EEG-based focus area localization with the proposed methods reaches 98.9% accuracy using the Rotation Forest classifier. Therefore, our results suggest that ensemble machine learning methods can be applied to differentiate the EEG signals from epileptogenic brain areas and signals recorded from non-epileptogenic brain regions with high accuracy.

Over the course of the last decade, the subfield of artificial intelligence, called deep learning, becomes the main technology that provides breakthroughs in the computer vision area. Likewise, deep learning algorithms made a major impact in the automated driving domain. This research aims to apply and evaluate the performance of two pre-trained deep learning algorithms in order to recognize different street objects. Both RCNN, as well as YOLO algorithms, are used to recognize bikes, cars and pedestrians using the public GRAZ-02 dataset composed of 1476 raw images of street objects. Accuracy greater than 90% is achieved in recognizing all considered objects. The fine-tuning and training of both algorithms is established using databases named ImageNet and COCO, and afterwards, trained models are tried on the test data.

Noninvasive load monitoring have been investigated by researchers for decades due to its cost-effective benefits. Upon introduction of smart meters, obtaining data about power consumption of households became easier. Numerous different techniques have been applied on the power consumption data to gain useful information out of it. This study applies machine learning techniques (Bayes network, random forest and rotational forest) to determine the operation state of households, where households are assumed to be either in ON or OFF state. Tracebase power consumption signature repository was used to train and test proposed machine learning models. Tracebase dataset was preprocessed to generate 4 different datasets. Test results have shown that these machine learning algorithms are able to estimate operation state with high accuracy and Bayes network shows outstanding performance among them with overall accuracy of 95%. Proposed method is extremely cost-effective for load monitoring and could replace some of the physical sensors in the smart houses.

This study presents a short-term prediction approach for honey production using ensemble regression technique. The data were recorded as a part of Habeetat project in Sarajevo, Bosnia and Herzegovina for 2016 season. This season has been entitled as one of the worst seasons for beekeepers in our country, which makes the problem of honey production prediction even more challenging. Random Tree regression algorithm was used for such purpose showing that the mean absolute error in predicting total honey production was less than 1.16 kg in all three hives monitored between November 2016 and April 2017. These findings are very significant for beekeepers since they can be notified in advance to visit individual hives and collect the honey. Besides, they can monitor trends in honey production throughout the season and perhaps change the position of hives in the current season and for the next upcoming season.