Data mining is most commonly used in attempts to induce association rules from transaction data. Most previous studies focused on binary-valued transaction data. Transaction data in real-world applications, however, usually consist of quantitative values. This paper, thus, proposes a fuzzy data-mining algorithm for extracting both association rules and membership functions from quantitative transactions. A genetic algorithm (GA)-based framework for finding membership functions suitable for mining problems is proposed. The fitness of each set of membership functions is evaluated by the fuzzy-supports of the linguistic terms in the large 1-itemsets and by the suitability of the derived membership functions. The evaluation by the fuzzy supports of large 1-itemsets is much faster than that when considering all itemsets or interesting association rules. It can also help divide-and-conquer the derivation process of the membership functions for different items. The proposed GA framework, thus, maintains multiple populations, each for one item's membership functions. The final best sets of membership functions in all the populations are then gathered together to be used for mining fuzzy association rules. Experiments are conducted to analyze different fitness functions and set different fitness functions and setting different supports and confidences. Experiments are also conducted to compare the proposed algorithm, the one with uniform fuzzy partition, and the existing one without divide-and-conquer, with results validating the performance of the proposed algorithm.

Privacy-preserving data mining (PPDM) has become an interesting and emerging topic in recent years because it helps hide confidential information, while allowing useful knowledge to be discovered at the same time. Data sanitization is a common way to perturb a database, and thus sensitive or confidential information can be hidden. PPDM is not a trivial task and can be concerned an Non-deterministic Polynomial-time (NP)-hard problem. Many algorithms have been studied to derive optimal solutions using the evolutionary process, although most are based on straightforward or single-objective methods used to discover the candidate transactions/items for sanitization. In this paper, we present a multi-objective algorithm using a grid-based method (called GMPSO) to find optimal solutions as candidates for sanitization. The designed GMPSO uses two strategies for updating gbest and pbest during the evolutionary process. Moreover, the pre-large concept is adapted herein to speed up the evolutionary process, and thus multiple database scans during each evolutionary process can be reduced. From the designed GMPSO, multiple Pareto solutions rather than single-objective algorithms can be derived based on Pareto dominance. In addition, the side effects of the sanitization process can be significantly reduced. Experiments have shown that the designed GMPSO achieves better side effects than the previous single-objective algorithm and the NSGA-II-based approach, and the pre-large concept can also help with speeding up the computational cost compared to the NSGA-II-based algorithm.

To determine an appropriate trading time for buying or selling stocks is always a difficult task. The common way to deal with it is using trading strategies formed by technical or fundamental indictors. Lots of approaches have been presented on how to form trading strategies and how to set suitable parameters for those strategies. Furthermore, some approaches were also designed to optimize a trading strategy portfolio, which is a set of strategies where the return and risk of the portfolio can be maximized and minimized, respectively. To provide a more useful trading strategy portfolio, we first define a group trading strategy portfolio (GTSP). Then, an algorithm that utilizes the grouping genetic algorithm is designed for solving the GTSP optimization problem. In the chromosome representation, the grouping, strategy and weight parts are employed to encode a possible GTSP. The fitness value of a chromosome is calculated by the group balance, weight balance, portfolio return, and risk to assess the quality of every possible solution. Genetic operators, including crossover, mutation and inversion, are applied on the population to form new offspring. Evolution is continued until the stop conditions are reached. At last, experiments were conducted on two real datasets with different trends to show the advantages and effectiveness of the presented approach.

Internet of Things (IoT) supports high flexibility and convenience in several applications because the IoT devices continuously transfer, share, and exchange data without human intervention. During shared or exchanged progress of data, security and privacy threats result because the published or shared data mainly corresponds to a raw dataset, and an attacker can easily obtain details on the shared data in an IoT environment. In the study, we present a sanitization approach by adopting the hierarchical-cluster method to hide confidential information while still discovering useful and meaningful information in the sanitized dataset. The multi-objective particle swarm optimization framework and an algorithm termed as HCMPSO are utilized to balance four side effects, namely hiding failure, missing cost, artificial cost, and database dissimilarity (Dis), and thereby provide optimized solutions for data sanitization. Experiments are performed to compare the performance of the designed HCMPSO with that of the single-objective cpGA2DT and multi-objective NSGA-II-based approaches. As shown in the results, the designed HCMPSO exhibits good performance in terms of hiding failure, and thus the most confidential information is hidden after the sanitization process. The shared or published data in IoT is secured. Furthermore, the designed sanitization algorithm achieves reasonable results in terms of missing cost, artificial cost, and Dis.