VIEWS: 85 PAGES: 8 CATEGORY: Emerging Technologies POSTED ON: 2/15/2011 Public Domain
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No. 1, January 2011 Finding Fuzzy Locally Frequent Itemsets Fokrul Alom Mazarbhuiya Md. Ekramul Hamid Department of Computer Science Department of Network Engineering College of Computer Science College of Computer Science King Khalid University King Khalid University Abha, Kingdom of Saudi Arabia Abha, Kingdom of Saudi Arabia e-mail: fokrul_2005@yahoo.com e-mail: ekram_hamid@yahoo.com Abstract— The problem of mining temporal association rules time stamps are having similar membership functions, we from temporal dataset is to find association between items that claim that the same algorithm with slight modification can be hold within certain time intervals but not throughout the dataset. applied to the database having dissimilar fuzzy time stamps. This involves finding frequent sets that are frequent at certain time intervals and then association rules among the items present In section II we give a brief discussion on the works related in the frequent sets. In fuzzy temporal datasets as the time of to our work. In section III we describe the definitions, terms transaction is imprecise, we may find set of items that are and notations used in this paper. In section IV, we give the frequent in certain fuzzy time intervals. We call these as fuzzy algorithm proposed in this paper for mining fuzzy locally locally frequent sets and the corresponding associated association frequent sets. In section V, we explain the algorithm with a rules as fuzzy local association rules. These association rules small dataset and display the results. We conclude with cannot be discovered in the usual way because of fuzziness conclusion and lines for future work in section VI. In the last involved in temporal features. In this paper, we propose a section we give some references. modification to the well-known A-priori algorithm to compute fuzzy locally frequent sets. Finally we have shown manually with II. RELATED WORKS the help of an example that the algorithm works. The problem of discovery of association rules was first Keywords- Temporal Data mining, Fuzzy number, Fuzzy time- formulated by Agrawal et al in 1993. Given a set I, of items stamp, Core length of a fuzzy interval, Fuzzified interval and a large collection D of transactions involving the items, the problem is to find relationships among the items i.e. the I. INTRODUCTION presence of various items in the transactions. A transaction The problem of mining association rules has been defined t is said to support an item if that item is present in t. A initially [15] by R. Agarwal et al for application in large super transaction t is said to support an itemset if t supports each markets. Large supermarkets have large collection of records of of the items present in the itemset. An association rule is daily sales. Analyzing the buying patterns of the buyers will an expression of the form X ⇒ Y where X and Y are subsets help in taking typical business decisions such as what to put on of the itemset I. The rule holds with confidence τ if τ% of sale, how to put the materials on the shelves, how to plan for the transaction in D that supports X also supports Y. The future purchase etc. rule has support σ if σ% of the transactions supports X ∪ Y. Mining for association rules between items in temporal A method for the discovery of association rules was given databases has been described as an important data-mining in [15], which is known as the A priori algorithm. This was problem. Transaction data are normally temporal. The market then followed by subsequent refinements, generalizations, basket transaction is an example of this type. extensions and improvements. As the number of association rules generated is too large, attempts were made to extract In this paper we consider datasets, which are fuzzy temporal i.e. the time in which a transaction has taken place is the useful rules ([13], [16]) from the large set of discovered imprecise or approximate and is attached to the transactions. In association rules. Attempts are also made to make the large volumes of such data, some hidden information or process of discovery of rules faster ([12], [14]). relation ship among the items may be there which cannot be Generalized association rules ([9], [17]) and Quantitative extracted because of some fuzziness in the temporal features. association rules ([18]) were later on defined and Also the case may be that some association rules may hold in algorithms were developed for the discovery of these rules. certain fuzzy time period but not throughout the dataset. For A hashed based technique is used in [11] to improve the finding such association rules we need to find itemsets that are rule mining process of the A priori algorithm. frequent at certain time period, which will obviously be Temporal Data Mining is now an important extension imprecise due to the fact that the time of each transaction is of conventional data mining and has recently been able to fuzzy. We call such frequent sets fuzzy locally frequent over attract more people to work in this area. By taking into fuzzy time interval. From these fuzzy locally frequent sets, account the time aspect, more interesting patterns that are associations among the items can be obtained. It is shown time dependent can be extracted. There are mainly two manually with the help of an example that the algorithm gives broad directions of temporal data mining [7]. One concerns the required result. Although it is assumed here that the fuzzy the discovery of causal relationships among temporally 139 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No. 1, January 2011 oriented events. Ordered events form sequences and the the previous paragraph similarly works were also done in cause of an event always occur before it. The other [23], [24] but in non-fuzzy temporal data. But in all these concerns the discovery of similar patterns within the same methods they discuss the association rule mining of non- time sequence or among different time sequences. The fuzzy temporal data. Our approach although little bit underlying problem is to find frequent sequential patterns similar to the work of [1], is different from others in the in the temporal databases. The name sequence mining is sense that it discovers association rules from fuzzy normally used for the underlying problem. In [8] the temporal data and finds the association rules along with problem of recognizing frequent episodes in an event their fuzzy time intervals over which the rules hold sequence is discussed where an episode is defined as a automatically. collection of events that occur during time intervals of a specific size. III. PROBLEM DEFINITION The association rule discovery process is also extended to incorporate temporal aspects. In temporal association A. Some Definition related to Fuzziness rules each rule has associated with it a time interval in which the rule holds. The problems associated are to find Let E be the universe of discourse. A fuzzy set A in E is valid time periods during which association rules hold, the characterized by a membership function A(x) lying in [0,1]. discovery of possible periodicities that association rules A(x) for x ∈ E represents the grade of membership of x in A. have and the discovery of association rules with temporal Thus a fuzzy set A is defined as features. In [10], [19], [20] and [21], the problem of A={(x, A(x)), x ∈ E } temporal data mining is addressed and techniques and algorithms have been developed for this. In [10] an A fuzzy set A is said to be normal if A(x) =1 for at least one algorithm for the discovery of temporal association rules is x∈E described. In [2], two algorithms are proposed for the An α-cut of a fuzzy set is an ordinary set of elements with discovery of temporal rules that display regular cyclic membership grade greater than or equal to a threshold α, 0 ≤ α variations where the time interval is specified by user to ≤ 1. Thus a α-cut Aα of a fuzzy set A is characterized by divide the data into disjoint segments like months, weeks, days etc. Similar works were done in [6] and [22] Aα={x ∈E; A(x) ≥α} [see e.g. [4]] incorporating multiple granularities of time intervals (e.g. A fuzzy set is said to be convex if all its α-cuts are convex first working day of every month) from which both cyclic sets [see e.g. [5]]. and user defined calendar patterns can be achieved. In [1], the method of finding locally and periodically frequent sets A fuzzy number is a convex normalized fuzzy set A defined and periodic association rules are discussed which is an on the real line R such that improvement of other methods in the sense that it 1. there exists an x0 ∈ R such that A(x0) =1, and dynamically extract all the rules along with the intervals where the rules hold. In ([23], [24]) fuzzy calendric data 2. A(x) is piecewise continuous. mining and fuzzy temporal data mining is discussed where A fuzzy number is denoted by [a, b, c] with a < b < c user specified ill-defined fuzzy temporal and calendric where A(a) = A(c) = 0 and A(b) = 1. A(x) for all x ∈[a, b] is patterns are extracted from temporal data. known as left reference function and A(x) for x ∈ [b, c] is Our approach is different from the above approaches. known as the right reference function. Thus a fuzzy number We are considering the fact that the time of transactions are can be thought of as containing the real numbers within some not precise rather they are fuzzy numbers and some items interval to varying degrees. The α-cut of the fuzzy number [t1- are seasonal or appear frequently in the transactions for a, t1, t1+a] is a closed interval [t1+(α-1).a, t1+(1-α).a]. certain ill-defined periods only i.e. summer, winter, etc. They appear in the transactions for a short time and then Fuzzy intervals are special fuzzy numbers satisfying the disappear for a long time. After this they may again following. reappear for a certain period and this process may repeat. 1. there exists an interval [a, b]⊂ R such that A(x0) =1 For these itemsets the support cannot be calculated in the for all x0∈ [a, b], and usual way ([1], [10]), it has to be computed by the method defined in section 3B. These items may lead to interesting 2. A(x) is piecewise continuous. association rules over fuzzy time intervals. In this paper A fuzzy interval can be thought of as a fuzzy number with a we calculate the support values of these sets locally in a α- flat region. A fuzzy interval A is denoted by A = [a, b, c, d] cut of a fuzzy time interval where a fuzzy time interval with a < b < c < d where A(a) = A(d) = 0 and A(x) = 1 for all represents a particular season in which the itemset is x ∈[b, c]. A(x) for all x ∈[a, b] is known as left reference appearing frequently and if they are frequent in the fuzzy function and A(x) for x ∈ [c, d] is known as the right reference time interval under consideration then we call these sets function. The left reference function is non-decreasing and the fuzzy locally frequent sets. The large fuzzy time gap in right reference function is non-increasing [see e.g. [3]]. which they do not appear is not counted. As mentioned in 140 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No. 1, January 2011 Similarly the α-cut of the fuzzy interval [t1-a, t1, t2, t2+a] is [t1-a, t1, t2, t2+a]. We say that an association rule X ⇒ Y, where a closed interval [t1+(α-1).a, t2+(1-α).a]. X and Y are item sets holds in the time interval [t1-a, t1, t2, t2+a] if and only if given threshold τ, The core of a fuzzy number A is the set of elements of A having membership value one i.e. Sup[t1 − a ,t1 ,t 2 ,t 2 + a ] ( X ∪ Y ) / Sup[t1 − a ,t1 ,t 2 ,t 2 + a ] ( X ) ≥ τ / 100.0 Core(A) = {(x, A(x); A(x) = 1} For every fuzzy set A, and X∪Y is frequent in [t1-a, t1, t2, t2+a]. In this case we say that the confidence of the rule is τ. U α α ∈[ 0 ,1] A For each locally frequent item set we keep a list of fuzzy A= time intervals in which the set is frequent where each fuzzy where αA(x) = α. αA(x), and αA is a special fuzzy set [4] interval is represented as [start-a, start, end, end+a] where start gives the approximate starting time of the time interval For any two fuzzy sets A and B and for all α∈[0, 1], and end gives the approximate ending time of the time-interval. i) α (A∪B) = αA ∪αB end – start gives the length of the core of the fuzzy time interval. For a given value of α of two intervals [start1-a, start1, α ii) (A∩B) = αA ∩αB end1, end1+a] and [start2-a, start2, end2, end2+a] are non- overlapping if their α-cuts are non-overlapping. For any two fuzzy numbers A and B, we say the membership functions A(x) and B(x) are similar to each other if IV. PROPOSED ALGORITHM the slope of the left reference function of A(x) is equal to the that of B(x) and the slope of right reference of A(x) is equal that A. Generating Fuzzy Locally Frequent Sets of B(x). Obviously for any two fuzzy numbers A and B having While constructing locally frequent sets, with each locally similar membership functions frequent set a list of fuzzy time-intervals is maintained in which ⏐ αA⏐ = ⏐αB⏐, ∀α∈[0, 1] the set is frequent. Two user’s specified thresholds α and minthd are used for this. During the execution of the algorithm while making a pass through the database, if for a particular B. Some Definition related to Fuzzy Locally Frequent set itemset the α-cut of its current fuzzy time-stamp, [αLcurrent, α Rcurrent] and the α-cut, [αLlastseen, αRlastseen] of its fuzzy Let T = <to, t1,…………> be a sequence of imprecise or fuzzy time, when it was last seen overlap then the current transaction time stamps over which a linear ordering < is defined where ti is included in the current time-interval under consideration < tj means ti denotes the core of a fuzzy time which is earlier which is extended with replacement of αRlastseen by than the core of another fuzzy time stamp tj. For the sake of α Rcurrent; otherwise a new time-interval is started with convenience, we assume that all the fuzzy time stamps are α Lcurrent as the starting point. The support count of the item having similar membership functions. Let I denote a finite set set in the previous time interval is checked to see whether it is of items and the transaction database D is a collection of frequent in that interval or not and if it is so then it is fuzzified transactions where each transaction has a part which is a subset and added to the list maintained for that set. Also for the fuzzzy of the itemset I and the other part is a fuzzy time-stamp locally frequent sets over fuzzy time intervals, a minimum core indicating the approximate time in which the transaction had length of the fuzzy period is given by the user as minthd and taken place. We assume that D is ordered in the ascending fuzzy time intervals of core length greater than or equal to this order of the core of fuzzy time stamps. For fuzzy time intervals value are only kept. If minthd is not used than an item we always consider a fuzzy closed intervals of the form [t1-a t1, appearing once in the whole database will also become locally t2, t2+a] for some real number a. We say that a transaction is in frequent a over fuzzy point of time. the fuzzy time interval [t1-a, t1, t2, t2+a] if the α-cut of the fuzzy time stamp of the transaction is contained in α-cut of [t1-a, t1, Procedure to compute L1, the set of all fuzzy locally frequent t2, t2+a] for some user’s specified value of α. item sets of size 1. We define the local support of an itemset in a fuzzy time For each item while going through the database we always interval [t1-a, t1, t2, t2+a] as the ratio of the number of keeps an α-cut αlastseen which is [αLlastseen, αRlastseen] that transactions in the time interval [t1+(α-1).a, t2+(1-α).a] corresponds to the fuzzy time stamp when the item was last containing the itemset to the total number of transactions in seen. When an item is found in a transaction and the fuzzy [t1+(α-1).a, t2+(1-α).a] for the whole data base D for a given time-stamp is tm and if its α-cut αtm=[αLtm, αRtm] has empty intersection with [αLlastseen, αRlastseen], then a new time Sup [ t1 − a , t1 , t 2 , t 2 + a ] value of α. We use the notation (X) to interval is started by setting start of the new time interval as α denote the support of the itemset X in the fuzzy time interval Ltm and end of the previous time interval as αRlastseen. The [t1-a, t1, t2, t2+a]. Given a threshold σ we say that an itemset X previous time interval is fuzzified provided the support of the is frequent in the fuzzy time interval [t1-a, t1, t2, t2+a] if item is greater than min-sup. The fuzzified interval is then Sup[t1 − a ,t1 ,t 2 ,t 2 + a ] added to the list maintained for that item provided that the (X) ≥ (σ/100)* tc where tc denotes the total duration of the core is greater than minthd. Otherwise α number of transactions in D that are in the fuzzy time interval Rlastseen is set to αRtm, the counters maintained for counting 141 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No. 1, January 2011 transactions are increased appropriately and the process is { if (itemcount[k]/transcount[k]*100≥σ) continued. fuzzify([αLlastseen[k], αRlastseen[k]], ∀α∈[0, 1]) Following is the algorithm to compute L1, the list of locally frequent sets of size 1. Suppose the number of items in the if(|core(fuzzified interval)|≥ minthd) dataset under consideration is n and we assume an ordering add(fuzzified interval) to tp[k]; among the items. if(tp[k] != 0) add {ik, tp[k]} to L1 • Algorithm1 } C1 = {(ik,tp[k]) : k = 1,2,…..,n} fuzzify([αa,αb], α) where ik is the k-th item and tp[k] points to a list of fuzzy time intervals initially empty.} U α α ∈[ 0 ,1] [ a, b ] for k = 1 to n do { fuzzified interval= ; α set lastseen[k]=φ; α where α[a, b](x) = α. [a, b](x) set itemcount[k]and transcount[k] to zero for each return(fuzzified interval) transaction t in the database with fuzzy time stamp tm } do Two support counts are kept, itemcount and transcount. If {for k = 1 to n do the count percentage of an item in an α-cut of a fuzzy time { if {ik} ⊆ t then interval is greater than the minimum threshold then only the set is considered as a locally frequent set over fuzzy time interval. { if(αlastseen[k] == φ) L1 as computed above will contain all 1-sized locally {αlastseen[k] = αfirstseen[k] = αtm; frequent sets over fuzzy time intervals and with each set there is associated an ordered list of fuzzy time intervals in which the itemcount[k] = transcount[k] = 1; set is frequent. Then A priori candidate generation algorithm is } used to find candidate frequent set of size 2. With each candidate frequent set of size two we associate a list of fuzzy else time intervals that are obtained in the pruning phase. In the if([αLlastseen[k],αRlastseen[k]]∩ generation phase this list is empty. If all subsets of a candidate [ Ltm[k], αRtm[k]]≠φ) α set are found in the previous level then this set is constructed. The process is that when the first subset appearing in the {αRlastseen[k]=αRtm[k]; itemcount[k]++; previous level is found then that list is taken as the list of fuzzy time intervals associated with the set. When subsequent subsets transcount[k]++; are found then the list is reconstructed by taking all possible } pair wise intersection of subsets one from each list. Sets for which this list is empty are further pruned. else Using this concept we describe below the modified A- { if (itemcount[k]/transcount[k]*100 ≥ σ) priori algorithm for the problem under consideration. fuzzify([αLlastseen[k],αRlastseen[k]],∀α • Algorithm2 ∈[0, 1]) Modified A priori if(|core(fuzzified interval)|≥ minthd) Initialize add(fuzzified interval) to tp[k]; k = 1; itemcount[k] = transcount[k] = 1; C1 = all item sets of size 1 lastseen[k] = firstseen[k] = tm; L1 = {frequent item sets of size 1 where } with each itemset {ik} a list tp[k] is maintained which } gives all time fuzzy else transcount[k]++; intervals in which the set is frequent} } // end of k-loop // L1 is computed using algorithm 1.1 */ } // end of do loop // for(k = 2; Lk-1 ≠ φ ; k++) do for k = 1 to n do { Ck = apriorigen(Lk-1) 142 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No. 1, January 2011 /* same as the candidate generation method of the A priori Fuzzy time-stamps set of transactions algorithm setting tp[i] to zero for all i*/ [0, 2, 4] 1 3 4 5 prune(Ck); [1, 3, 5] 1 4 6 7 9 [2, 4, 6] 2 3 5 6 7 10 drop all lists of fuzzy time intervals maintained with the [3, 5, 7] 1 3 5 7 10 sets in Ck [4, 6, 8] 2 3 4 8 9 Compute Lk from Ck. [5, 7, 9] 1 2 4 5 6 10 [6, 8, 10] 1 2 3 4 7 8 10 //Lk can be computed from Ck using the same procedure [7, 9, 11] 1 2 3 4 5 6 7 8 used for computing L1 // [8, 10, 12] 1 2 3 4 5 7 8 9 k=k+1 [9, 11, 13] 2 3 4 6 7 10 [10, 12, 14] 1 2 3 4 5 7 10 } [11, 13, 15] 1 2 8 10 [12, 14, 16] 1 2 3 4 5 7 UL k k [13, 15, 17] 2 3 6 7 Answer = [14, 16, 18] 1 2 3 4 5 Prune(Ck) [15, 17, 19] 1 2 3 4 [16, 18, 20] 2 5 6 7 {Let m be the number of sets in Ck and let the sets be s1, [17, 19, 21] 1 2 3 4 5 s2,…, sm. Initialize the pointers tp[i] pointing to the list of [18, 20, 22] 2 4 8 10 fuzzy time-intervals maintained with each set si to null [19, 21, 23] 2 3 6 9 for i = 1 to m do [20, 22, 24] 3 4 7 8 [21, 23, 1’] 1 6 10 {for each (k-1) subset d of si do [22, 24, 2’] 1 5 {if d ∉ Lk-1 then [23, 1’, 3’] 7 [24, 2’, 4’] 2 10 {Ck = Ck - {si, tp[i]}; break;} [1’, 3’, 5’] 3 4 else [2’, 4’, 6’] 1 4 5 8 9 10 { if (tp[i] == null) then set tp[i] to point to the list of [3’, 5’, 7’] 1 2 4 fuzzy time intervals maintained for d [4’, 6’, 8’] 2 4 5 [5’, 7’, 9’] 2 3 4 6 7 else [6’, 8’, 10’] 3 5 { take all possible pair-wise intersection of fuzzy time [7’, 9’, 11’] 1 3 4 5 6 7 intervals one from each list,one list maintained with tp[i] and [8’, 10’, 12’] 2 5 the other maintained with d and take this as the list for tp[i] [9’, 11’, 13’] 1 2 3 7 8 [10’, 12’, 14’] 1 9 10 delete all fuzzy time intervals whose core length is less than [11’, 13’, 15’] 1 2 3 6 10 the value of minthd if tp[i] is empty then {Ck = Ck - [12’, 14’, 16’] 2 3 4 {si,tp[i]}; [13’, 15’, 17’] 1 2 3 break; [14’, 16’, 18’] 1 2 3 [15’, 17’, 19’] 3 4 5 7 8 } [16’, 18’, 20’] 1 2 3 8 10 } [17’, 19’, 21’] 2 3 5 [18’, 20’, 22’] 1 2 3 4 5 6 } [19’, 21’, 23’] 1 2 3 5 7 8 } [20’, 22’, 24’] 2 3 4 5 9 10 } We execute the algorithm manually with the above dataset } taking min-sup = 0.4 and minthd = 3. After first pass we have the set of 1-item frequent sets along V. EXPLANATION OF THE ALGORITHM WITH EXAMPLE with the fuzzy intervals where they are frequent as To illustrate the above algorithms, we consider a dataset of L1={({1}; [0, 2, 19, 21], [7’, 9’, 21’, 23’]), two days consisting of fuzzy time stamps and set of ({2}; [2, 4, 21, 23], [8’, 10’, 22’, 24’]), transactions. Here each fuzzy time stamp is associated with a ({3}; [0, 2, 22, 24], [5’, 7’, 22’, 24’]), transactions means that the transaction occurs at a fuzzy time. ({4}; [4, 6, 22, 24], [1’, 3’, 9’, 11’]), For the sake of convenience, we take all the fuzzy time stamps ({5}; [0, 2, 19, 21], [2’, 4’, 10’, 12’], as triangular fuzzy numbers. The dataset is given below: [15’, 17’, 22’, 24’]), 143 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No. 1, January 2011 ({6}; [5, 7, 11, 13]), ({3 4 8}; [4, 6, 10, 12]), ({7}; [6, 8, 15, 17], [5’, 7’, 11’, 13’]), ({3 5 7}; [7, 9, 14, 16]), ({8}; [4, 6, 10, 12]), ({4 5 7}; [7, 9, 14,16])} ({10}; [2, 4, 8, 10])} Candidates for the fourth pass are Candidates for the second pass are C2={{1 2}, {1 3}, {1 4}, {1 C4 = {{1 2 3 4}, {1 2 3 5}, {1 2 3 7}, {1 2 4 5}, {1 2 4 7}, {1 5}, {1 6}, {1 7}, {1 8}, {1 10}, {2 3}, {2 4}, {2 5}, {2 6}, {2 2 5 7}, {1 3 4 5}, {1 3 4 7}, {1 3 5 7}, {2 3 4 5}, {2 3 4 7}, {2 7}, {2 8}, {2 10}, {3 4}, {3 5}, {3 6}, {3 7}, {3 8}, {3 10}, 3 4 8}, {2 3 5 7}, {2 4 5 7}, {2 4 5 7}} {4, 5}, {4 6}, {4 7}, {4 8}, {4 10}, {5 6}, {5 7}, {5 8}, {5 L4={({1 2 3 4}; [6, 8, 19, 21]), 10}, {6 7}, {6 8}} ({1 2 3 5}; [6, 8, 16, 18]), After the second pass, we got the second level frequent sets as ({1 2 3 7}; [6, 8, 14, 16]), L2={({1 2};[5, 7, 19, 21],[9’, 11’, 21’, 23’]), ({1 2 4 5}; [5, 7, 16, 18]), ({1 3}; [6, 8, 19, 21], [7’, 9’, 21’, 23’]), ({1 2 4 7}; [6, 8, 14, 16]), ({1 4}; [5, 7, 19, 21]), ({1 2 5 7}; [6, 8, 14, 16]), ({1 5}; [3, 5, 16, 18]), ({1 3 4 5}; [7, 9, 16, 18]), ({1 7}; [6, 8, 14, 16]), ({1 3 4 7}, [6, 8, 14, 16]), ({1 10}; [3, 5, 8, 10]), ({1 3 5 7}; [7, 9, 14, 16]), ({2 3}; [2, 4, 21, 23], [9’, 11’, 22’, 24’]), ({2 3 4 5}; [7, 9, 16, 18]), ({2 4}; [4, 6, 20, 22]), ({2 3 4 7}; [6, 8, 15, 17]), ({2 5}; [7, 9, 19, 21], [17’, 19’, 22’, 24’]), ({2 3 4 8}; [4, 6, 10, 12]), ({2 6}; [5, 7, 11, 13]), ({2 3 5 7}; [7, 9, 15, 17]), ({2 7}; [6, 8, 15, 17]), ({2 4 5 7}; [6, 8, 14, 16]), ({2 8}; [4, 6, 10, 12]), ({3 4 5 7}; [7, 9, 12, 14])} ({3 4}; [4, 6, 19, 21]), Candidates for the fifth pass are ({3 5}; [0, 2, 5, 7], [7, 9, 16, 18], C5 ={{1 2 3 4 5}, {1 2 3 4 7}, {1 2 3 5 7}, {1 2 4 5 7}, {1 3 4 [15’, 17’, 22’, 24’]), 5 7}, {2 3 4 5 7}} ({3 7}; [6, 8, 15, 17], [5’, 7’, 11’, 13’]), After the fifth pass, we got fifth level frequent sets as ({3 8}; [4, 6, 10, 12]), L5 = {({1 2 3 4 5}; [7, 9, 16, 18]), ({4 5}; [5, 7, 16, 18]), ({1 2 3 4 7}; [6, 8, 14, 16]), ({4 7}; [6, 8, 14, 16]), ({1 2 3 5 7}; [7, 9, 14, 16]), ({4 8}; [4, 6, 10, 12]), ({1 2 4 5 7}; [7, 9, 14, 16]), ({5 7}; [7, 9, 14, 16]), ({1 3 4 5 7}; [7, 9, 14, 16]), ({5 10}; [2, 4, 7, 9])} ({2 3 4 5 7}; [7, 9, 14, 16])} Candidates for the third pass are Candidates for sixth pass are C6 = {{1 2 3 4 5 7}} C3={{1 2 3}, {1 2 4}, {1 2 5}, {1 2 7}, {1 3 4}, {1 3 5}, {1 3 After the sixth pass we got sixth level frequent sets as 7}, {1 4 7}, {1 5 7}, {2 3 4}, {2 3 5}, {2 3 7}, {2 3 8}, {2 4 L6 = {({1 2 3 4 5 7}; [7, 9, 14,16])} 5}, {2 4 7}, {2 4 8}, {2 5 7}, {3 4 5}, {3 4 7}, {3 4 8},{3 5 7}, Answer = {({1 2 3 4 5 7}; [7, 9,14, 16]), {4 5 7}, {4 5 8}} ({2 3 4 8}; [4, 6, 10, 12]), After the third pass, we got the third level frequent sets as ({1 2}; [9’, 11’, 21’, 23’]), L3={({1 2 3}; [6, 8, 19, 21], [9’, 11’, 21’, 23’]), ({1 3}; [7’, 9’, 21’, 23’]), ({1 2 4}; [5, 7, 19, 21]), ({1 10}; [3, 5, 8,10]), ({1 2 5}; [5, 7, 16, 18]), ({2 3}; [9’, 11’, 22’, 24’]), ({1 2 7}; [6, 8, 14, 16]), ({2 5}; [17’, 19’, 22’, 24’]), ({1 3 4}; [6, 8, 19, 21]), ({3 5}; [15’, 17’, 22’, 24’]), ({1 3 5}; [7, 9, 16, 18]), ({3 7}; [5’, 7’, 11’, 13’]), ({1 3 7}; [6, 8, 14, 16]), ({5 10}; [2, 4, 7, 9]), (1 4 7}; [6, 8, 14, 16]), ({4}; [1’, 3’, 9’, 11’]), ({1 5 7}; [7, 9, 14, 16]), ({5}; [2’, 4’, 10’, 12’])} ({2 3 4}; [4, 6, 19, 21]), ({2 3 5}; [7, 9, 16, 18]), B. Generating Association Rules ({2 3 7}; [6, 8, 15, 17]), If an itemset is frequent in a fuzzy time-interval [t1-a, t1, t2, ({2 3 8}; [4, 6, 10, 12]), t2+a] then all its subsets are also frequent in the fuzzy time- ({2 4 5}; [5, 7, 16, 18]), interval [t1-a, t1, t2, t2+a]. But to generate the association rules ({2 4 7}; [6, 8, 14, 16]), as defined in section 3, we need the supports of the subsets in ({2 4 8}; [4, 6, 10, 12]), fuzzy time-interval [t1-a, t1, t2, t2+a], which may not be ({2 5 7}; [7, 9, 14, 16]), available after application of the algorithm as defined in 4.1. ({3 4 5}; [7, 9, 16, 18]), For this one more scan of the whole database will be needed. ({3 4 7}; [6, 8, 14, 16]), 144 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No. 1, January 2011 For each association rule we attach a fuzzy interval in which International Conference on KDD and data mining (KDD ’97), Newport Beach, California, August 1997. the association rule holds. [13] M. Klemettinen, H. Manilla, P. Ronkainen, H. Toivonen and A. I. CONCLUSIONS AND LINES FOR FUTURE WORKS Verkamo; Finding interesting rules from large sets of discovered association rules; Proceedings of the 3RD international Conference on An algorithm for finding frequent sets that are frequent in Information and Knowledge Management, Gathersburg, Maryland, 29 certain fuzzy time periods from fuzzy temporal data, is given in Nov 1994. the paper. The algorithm dynamically computes the frequent [14] R. Agrawal and R. Srikant; Fast algorithms for mining association rules, sets along with their fuzzy time intervals where the sets are Proceedings of the 20th International Conference on Very Large Databases (VLDB ’94), Santiago, Chile, June 1994. frequent. These frequent sets are named as fuzzy locally [15] R. Agrawal, T. Imielinski and A. Swami; Mining association rules frequent setsl. The technique used is similar to the A priori between sets of items in large databases; Proceedings of the ACM algorithm. From these fuzzy locally frequent sets interesting SIGMOD ’93, Washington, USA, May 1993. rules may follow. [16] R. Motwani, E. Cohen, M. Datar, S. Fujiware, A. Gionis, P. Indyk, J. D. Ullman and C. Yang; Finding interesting association rules without In the level-wise generation of fuzzy locally frequent sets, support pruning, Proceedings of the16th International Conference on for each fuzzy locally frequent set we keep a list of all fuzzy Data Engineering (ICDE), IEEE, 2000. time-intervals in which it is frequent. For generating candidates [17] R. Srikant and R. Agrawal; Mining generalized association rules, for the next level, pair-wise intersections of the intervals in two Proceedings of the 21st Conference on very large databases (VLDB ’95), lists are taken. Further, we tested manually with an example Zurich, Switzerland, September 1995. that the algorithm works. For the sake convenience, we have [18] R. Srikant and R. Agrawal; Mining quantitative association rules in large taken here the dataset having fuzzy time stamps with similar relational tables, Proceedings of the 1996 ACM SIGMOD Conference on management of data, Montreal, Canada, June 1996. membership functions, but the algorithm can be applicable to [19] X. Chen and I. Petrounias; A framework for Temporal Data Mining; the dataset with dissimilar fuzzy time stamps. The same Proceedings of the 9th International Conference on Databases and Expert algorithm can be implemented with both real life as well as Systems Applications, DEXA ’98, Vienna, Austria. Springer-Verlag, synthetic datasets. Berlin; Lecture Notes in Computer Science 1460, 796-805, 1998. [20] X. Chen and I. Petrounias; Language support for Temporal Data Mining; REFERENCES Proceedings of 2nd European Symposium on Principles of Data Mining and Knowledge Discovery, PKDD ’98, Springer Verlag, Berlin, 282- 290, 1998. [1] A. K. Mahanta, F. A. Mazarbhuiya and H. K. Baruah; Finding Locally and Periodically Frequent Sets and Periodic Association Rules, [21] X. Chen, I. Petrounias and H. Healthfield; Discovering temporal Proceeding of 1st Int’l Conf on Pattern Recognition and Machine Association rules in temporal databases; Proceedings of IADT’98 Intelligence (PreMI’05),LNCS 3776, 576-582, 2005. (International Workshop on Issues and Applications of Database Technology, 312-319, 1998. [2] B. Ozden, S. Ramaswamy and A. Silberschatz; Cyclic Association Rules, Proc. of the 14th Int’l Conference on Data Engineering, USA, [22] Y. Li, P. Ning, X. S. Wang and S. Jajodia; Discovering Calendar-based 412-421, 1998. Temporal Association Rules, In Proc. of the 8th Int’l Symposium on Temporal Representation and Reasonong, 2001. [3] D. Dubois and H. Prade; Ranking fuzzy numbers in the setting of possibility theory, Information Science 30, 183-224, 1983. [23] W. J. Lee and S. J. Lee; Discovery of Fuzzy Temporal Association Rules, IEEE Transactions on Systems, Man and Cybenetics-part B; [4] G. J. Klir and B. Yuan; Fuzzy Sets and Fuzzy Logic Theory and Cybernetics, Vol 34, No. 6, 2330-2341, Dec 2004. Applications, Prentice Hall India Pvt. Ltd., 2002. [24] W. J. Lee and S. J. Lee; Fuzzy Calendar Algebra and It’s Applications to [5] G. Q. Chen, S. C. Lee and E. S. H. Yu; Application of fuzzy set theory Data Mining, Proceedings of the 11th International Symposium on to economics, In: Wang, P. P., ed., Advances in Fuzzy Sets, Possibility Temporal Representation and Reasoning (TIME’04), IEEE, 2004. Theory and Applications, Plenum Press, N. Y., 277-305, 1983. [6] G. Zimbrao, J. Moreira de Souza, V. Teixeira de Almeida and W. Araujo da Silva; An Algorithm to Discover Calendar-based Temporal Association Rules with Item’s Lifespan Restriction, Proc. of the 8th AUTHOR’S PROFILE ACM SIGKDD Int’l Conf. on Knowledge Discovery and Data Mining (2002) Canada, 2nd Workshop on Temporal Data Mining, v. 8, 701-706, 2002. Fokrul Alom Mazarbhuiya received B.Sc. [7] J. F. Roddick, M. Spillopoulou; A Biblography of Temporal, Spatial and degree in Mathematics from Assam University, Spatio-Temporal Data Mining Research; ACM SIGKDD, June 1999. India and M.Sc. degree in Mathematics from [8] H. Manilla, H. Toivonen and I. Verkamo; Discovering frequent episodes Aligarh Muslim University, India. After this he in sequences; KDD’95; AAAI, 210-215, August 1995. obtained the Ph.D. degree in Computer Science [9] J. Hipp, A. Myka, R. Wirth and U. Guntzer; A new algorithm for faster from Gauhati University, India. Since 2008 he mining of generalized association rules; Proceedings of the 2nd European has been serving as an Assistant Professor in College of Symposium on Principles of Data Mining and Knowledge Discovery (PKDD ’98), Nantes, France, (September 1998. Computer Science, King Khalid University, Abha, kingdom of [10] J. M. Ale and G.H. Rossi; An approach to discovering temporal Saudi Arabia. His research interest includes Data Mining, association rules; Proceedings of the 2000 ACM symposium on Applied Information security, Fuzzy Mathematics and Fuzzy logic Computing, March 2000. [11] J. S. Park, M. S. Chen and P. S. Yu; An Effective Hashed Based Algorithm for Mining Association Rules; Proceedings of ACM SIGMOD, 175-186, 1995. [12] M. J. Zaki, S. Parthasarathy, M. Ogihara and W. Li; New algorithms for the fast discovery of association rules; Proceedings of the 3rd 145 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No. 1, January 2011 Md. Ekramul Hamid received his B.Sc and M.Sc degree from the Department of Applied Physics and Electronics, Rajshahi University, Bangladesh. After that he obtained the Masters of Computer Science degree from Pune University, India. He received his PhD degree from Shizuoka University, Japan. During 1997-2000, he was a lecturer in the Department of Computer Science and Technology, Rajshahi University. Since 2007, he has been serving as an associate professor in the same department. He is currently working as an assistant professor in the college of computer science at King Khalid University, Abha, KSA. His research interests include speech enhancement, and speech signal processing. 146 http://sites.google.com/site/ijcsis/ ISSN 1947-5500