Hyperparameters and architecture greatly influence the performance of convolutional neural networks (CNNs); therefore, their optimization is important to obtain the desired results. One of the state-of-the-art methods to achieve this is the use of neuroevolution that utilizes a genetic algorithm (GA) to optimize a CNN. However, the GA is often trapped into a local optimum resulting in premature convergence. In this study, we propose an approach called the “diversity-guided genetic algorithm-convolutional neural network (DGGA-CNN)” that uses adaptive parameter control and random injection to facilitate the search process by exploration and exploitation while preserving the population diversity. The alternation between exploration and exploitation is guided by using an average pairwise Hamming distance. Moreover, the DGGA fully handles the architecture of the CNN by using a novel finite state machine (FSM) combined with three novel mutation mechanisms that are specifically created for architecture chromosomes. Tests conducted on suggestion mining and twitter airline datasets reveal that the DGGA-CNN performs well with valid architectures and a comparison with other methods demonstrates its capability and efficiency.

Sentiment analysis from social media has turned out to be essential since individuals are normally genuine with their sentiment on giving their perspective. However, in turning social media into a sentiment analysis possess challenges such as comments are usually ambiguous, language barrier problem, slang words, redundant comment, and sentiment classification. This study attempted to distinguish the issues of sentiment classification from Indonesian social media on Jakarta governor election. Several steps are taken to overcome those problems that include preprocessing. The preprocessing strategy used are removing the unrelated tweet, removing URL, deleting duplicate lines, deleting similar lines, removing the unrelated word, removing hashtag, removing Twitter username, removing number in the comment, removing punctuation, checking slang words, and converting the slang word into appropriate word. The preprocessed sentiment is then classified into positive, negative, and neutral. The classification method used in this study are Summation method, Average on Tweet, Average on Tweet with the threshold on objective score, Weighted Average, and Naïve Bayes method. The experimental results show that Naïve Bayes produce the highest precision, highest recall, and highest accuracy for neutral and positive sentiment. But, Naïve Bayes does not produce good results for negative sentiment.

The optimization problems on real-world usually have non-linear characteristics. Solving non-linear problems is time-consuming, thus heuristic approaches usually are being used to speed up the solution’s searching. Among of the heuristic-based algorithms, Genetic Algorithm (GA) and Simulated Annealing (SA) are two among most popular. The GA is powerful to get a nearly optimal solution on the broad searching area while SA is useful to looking for a solution in the narrow searching area. This study is comparing performance between GA, SA, and three types of Hybrid GA-SA to solve some non-linear optimization cases. The study shows that Hybrid GA-SA can enhance GA and SA to provide a better result

Research in sentiment analysis often have very high features for the classification and it might affect the model's accuracy. In this paper, the Variable Length Chromosome Genetic Algorithm with Naïve Bayes (VLCGA-NB) is utilized to analyze the twitter sentiment. The tweets are preprocessed in several steps before using it in the algorithm. The preprocessing conducted to reduce the number of features. After the preprocessing performed, the features that produce a higher fitness value is selected. There are five classes: Very Positive, Positive, Neutral, Negative, and Very Negative to be classified. Comparison of NB and VLCGA-NB is conducted to show the models accuracy. The experiments show that VLCGA-NB produces the higher value for the Best F-Measure results of each sentiment, lower number of features (188 features), higher accuracy (75.2%), and higher fitness value (3.088953) than NB.

Time series datasets collected from marine sensors inevitably undergo missing data problems. This cause unreliable sensor data to assist the decision-making process. Many methods are offered to impute missing values. However, selecting the best imputation method is not a trivial task, as it usually requires domain expertise and several trial-and-error iterations. Furthermore, when imputations are carried out in a careless way, it generates a high error factor that can lead stakeholders to wrong assumptions. This paper provides a systematic approach that is able to extract characteristics of underlying data and, based on it, recommends the less error-prone imputation method. We evaluate our proposed method using nine real-world vessel datasets. In total, we generated 3859 data samples consisting of 17 inputs and 1 target feature. Experimental results show that the proposed approach is capable of obtaining a weighted F1-Score of 92.6%. Additionally, when compared with the application of careless selected imputation methods, our work is able to gain up to 86% on the average imputation score, with the worst case gain being of 5%. We empirically demonstrate that the proposed approach is efficient when selecting the best imputation methods.