Research On Mining Of Prolific Players: Going Beyond Theconventional Measures

Analytical trends in sports have been shifted from conventional statistics to contemporary data mining analytics. Among leading sports, cricket analyt-ics are lagging far behind and demand the scholastic and experts attentions. To fill huge gaps as well as to address the limitations of current traditional statis-tics, data mining tools come forth as promising solution. The efficiency and effectiveness of data mining tools is no more hidden in data sciences. Until now, very few cricket concerned solutions by research community have been intended to take advantage of data mining tools, rather, most of the addressed works are utilizing orthodox statistics and optimization theory. The cricket or-ganizations are demanding worthy metrics and mechanisms that could incorpo-rate diversified insights, therefore, could be used for effective decision making.Moreover, such analytics would be highly beneficial for the authorities includ-ing coaches and managers for getting the optimal performances from players and teams while boosting their expertise. With this aim, the purpose of under-lying dissertation is to provide efficient and effective solutions of problems in cricket domain that are either not considered before or still having severe limi-tations to be addressed. More precisely, advanced data mining tools including supervised machine learning and random walk based algorithm are incorporat-ed for providing efficacious solutions about cricket analytics. In this disserta-tion, three key problems are addressed, which are the distinctive representative of cricket game and yield contemporary analytics:· An efficient methodology for rising stars prediction within the cricket do-main is presented while incorporating the concept of Co-players. A set of 9 features is formulated for rising stars prediction of batsmen as well as a set of 11 features is scrutinized for the bowlers. These features are never considered before. By testing different classification algorithms on our datasets, four appropriate machine learning classification algorithms including Bayesian Network, Naive Bayesian, Support Vector Machine,and Classification and Regression Tree are selected for binary classifica-tion of rising stars. Co-batsmen Runs for batsmen and Team Average for bowlers are found as leading individual features. While performing cate-gory wise analysis, it is revealed that Co-batsmen category for batsmen,while Team category for the bowlers outperforms the others. Further, it is discovered that generative classifiers outperforms the others. Rising stars prediction is made with high accuracy, and rankings of leading rising stars from both domains based on three defined metrics are presented and com-pared with the ICC rankings of players from 2013-2016. This innovative idea can be used for rising star prediction in other sports domains such as baseball, football and basketball etc.· An efficient mechanism for star cricketers prediction within the cricket domain is put forward while incorporating the concept of performance evolution. Sets of 8 and 7 features are formulated for mining star crick-eters from batting and bowling domains, respectively. By testing different classification algorithms on incorporated datasets, six appropriate super-vised machine learning classification algorithms including Simple Logis-tic, Support Vector Machines, Bayesian Network, Naive Bayesian, Ran-dom Forests and Classification and Regression Tree are selected for binary classification of star cricketers. During evaluations, 4 Runs Evolution for batsmen and Maiden Overs Evolution for bowlers are found as leading in-dividual features. While performing category wise analysis, it is revealed that Primary category for batsmen, while Derived category for the bowlers outperforms the other. Further, it is found that Bayesian rule based classi-fiers outperforms the others. Star cricketers prediction is made with high accuracy, and rankings of leading star cricketers from both domains based on performance evolution metric are presented and compared with the IC-C rankings of the cricketers from 2013-2016. The presented innovatory idea can be used for evolution based prediction of players in other sports domains such as football, basketball and baseball etc.. An effective mechanism is presented for the extraction of batting, bowling and team precedence within the cricket domain. Sets of 9 and 8 features are formulated for mining batting and bowling precedences, respective-ly. These features are aggregated to form productivity of weights of re-spective domains. Subsequently, the productivity weights of batting and bowling domains are combined to acquire team productivity weight. The employed features overcome the limitations of prevalent works in terms of missing insights and the proposed productivity weights were never con-sidered before in such appropriate manner. A novel Productivity Prece-dence Algorithm (PPA) is presented to mine batting, bowling and team precedence of each team against the others. By using the batting, bowl-ing and team productivity weights as inputs of PPA, span wise evolutions of underlying teams in terms of batting productivity precedence, bowling productivity precedence and team productivity precedence are extracted from January, 1971-January, 2017. During evaluations, it is discovered that Australian cricket team is dominating in all precedences. Moreover,the yielded team productivity precedence is cross checked against the win-ner of World Cup of respective spans, and it is found that it is not neces-sary the most productive team of respective span to win the World Cup.