PREDICTIVE SYSTEM BASED MULTI-LAYERED CLUSTERING MODEL AND LEAST ABSOLUTE SHRINKAGE AND SELECTION OPERATOR (LASSO)

• Main Author: Febianti, Fevi
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/36690
• Institution: Institut Teknologi Bandung

Companies that can grow rapidly are companies that are able to develop their analytical abilities in various part of business process, e.g. sales prediction process. Not so many companies have applied computation as an option to predict sales in the future, and chose to rely on the judgement of business analysts. In addition, several predictive methods that had been developed, rarely used computational complexity as one of the focuses of the research. This study is aimed to design a prediction system that not only provide good accuracy, but also reduce computational complexity. This study used multi-layered clustering model to model data. K-means++ is used as a clustering method in multi-layered clustering model because of its simplicity in weighting technique and efficiency in computation. Then this study also used least absolute shrinkage and selection operator (LASSO) method to extract the previous modeled data so that the best fit prediction model can be obtained. Experiments were conducted using sales data from a garment company in Indonesia. Performance of multi-layered clustering is measured using Dunn Validity Index (DVI), while performance of prediction is measured using Mean Absolute Shrinkage and Selection Operator (MAPE). The evaluation results indicate that the proposed prediction system can improve accuracy by 3% when compared with using only feature extraction as prediction method. Also, this prediction system can reduce computational complexity by 89.77% when compared with using only feature extraction.