A Genetic Algorithm Implementation oftheFuzzyLeastTrimmed Squares Clustering
John Benson Louis · 2007
Thispaperdescribes anewapproach tofinding a global solution forthefuzzyleast trimmedsquares clustering. Theleast trimmedsquares (LTS)estimator isknowntobea highbreakdown estimator, inbothregression andclustering. Fromthepoint ofviewofimplementation, thefeasible solution algorithm isoneofthefewknowntechniques thatguarantees a global solution fortheLTS estimator. Thefeasible solution algorithm divides a noisy datasetintotwoparts - thenon- noisy retained setandthenoisy trimmedset, byimplementing a pairwise swapofdatumbetween thetwosetsuntil a least squares estimator provides thebestfit ontheretained set.We present anovelgenetic algorithm-based implementation ofthe feasible solution algorithm forfuzzyleast trimmedsquares clustering, andalsosubstantiate theefficacy ofourmethodby three examples.