kasus baru - Generally : Berapa banyak perbedaan karakteristik/atribut Strategi Adaptasi tradisional - Reinstantiation solusi kasus hasil retrieve langsung dicopy menjadi solusi dari kasus baru tanpa modifikasi - Subtitution update solusi yg sesuai dgn kasus - Transformation solusi tdk boleh kontradiksi atau bertolak belakang dgn atributnya. Dibutuhkan pakar. Metode Adaption - Learning adaption solusi kasus lama dipelajari dahulu - Integrating Ruleand Case-Base d Adaptation Approa ches - Using an Adaptation Matrix fitur dlm kasus saling ketergantungan (ex : antara harga dan merk/produk) - Using Configuration Techniques Adaptasi kasus dengan machine learning - Fuzzy Decision tree (ex: ID3) Step1. Learn the feature weights of the problem attributes. 2. Cluster the cases using the weighted distances. 3. For every case in each cluster, compare the differences between this case and other cases in both problem attributes and solution attributes. 4. Fuzzify the discrepancy vectors. 5. For every case, a fuzzy decision tree is obtained through training using the fuzzy discrepancy vectors. Generate the fuzzy decision rules from the tree, and simplify the rules using the rough set technique (i.e., reducts). 6. Given a new problem, retrieve the most similar case from the case base. Fuzzify the
difference between these two cases in the
problem parts. Use the fuzzified difference as input to the fuzzy decision tree. 7. Use the adjustment values to adapt the old solution. - Backpropagation Neural Network Step 1,2,3 sama kayak Fuzzy Decision tree. 4. Preprocess (e.g. normalize) the discrepancy vectors for training the neural network(nn). 5. For each case, use its discrepancy vectors to train the nn. 6. retrieve the most similar case from the casebase. Compute the difference between their problem attributes. Preprocess this difference and use it as the input to the nn. 7. Obtain the adjustment amount from the nn by inputting the preprocessed differences of the attribute values. 8. Postprocess (e.g. denormalize) the adjustment amount and use it to adapt the old solution for the query problem. - Bayesian Model (BM) the most similar case is retrieved and the difference between their problem parts is input to the BM. The output is the class label that identifies the adjustment strategy Step 1,2,3 sama kayak Fuzzy Decision tree. 4. Fuzzify the discrepancy vectors. Estimate the parameters of the Bayesian model. 5. Given a new problem, retrieve the most similar case from the case base. Fuzzify the difference between these two cases in the problem parts. 6. Input the problem difference into the Byesian classifier. Identify the clas label for adjustment using the Byes formula. 7. Obtain d adjustmnt in d form of a linguistic vriable wth certainty fctors. Defuzzify thm to get the adjustment amount.
8. Use the adjustment amounts to adapt the old
solution. - Support Vector Machine Step 1,2,3 sama kayak Fuzzy Decision tree. 4. For each case, construct the support vector machine using the discrepancy vectors. 5. Given a new query, find its most similar case and compare the discrepancy in the problem part. 6. Input X0 and output the label corresponding to y0 to determine the adaptation strategy. 7. Use the adaptation strategy to derive a new solution. - Genetic algorithm The idea is that whe n an old solution is retrieved, a modification process is initialized randomly. The solution adapt ed is tested and the feedback is collected to determine its fitness. This process is repeated for many cycles until a satisfactory solution is ob tained. CBR VS Prob VS SPK CBR adalah suatu cara penyelesaian masalah baru dengan cara menalar dengan menggunakan kasuskasus lama. SPK adl suatu sistem yg memiliki prosedurprosedur pemroses data yang digunakan untuk membantu pengambilan keputusan. CBR berbeda dengan SPK, krn CBR adalah salah satu cara/prosedur untuk pengambilan keputusan dalam sistem pendukung keputusan. (Pengambilan keputusan untuk suatu masalah baru menggunakan kasuskasus lama yang sudah pernah terjadi sebelumnya) Probabilistic reasoning juga dapat diterapkan dalam sistem pendukung keputusan, pengambilan keputusan untuk masalah baru menggunakan probabilitas, misalnya conditional probability (peluang sebuah event terjadi karena event lainnya)