DEVELOPMENT OF A MULTIVARIATE HIGH- ORDER FUZZY TIME SERIES FORECASTING MODEL WITH DATA CLUSTERING FOR OPTIMUM PREDICTION AND CONTROL OF HANDOVER-BASED MOBILITY MANAGEMENT

(A CASE STUDY OF AIRTEL LAGOS)

By

Author

Presented To

Department of Engineering

ABSTRACT
The research is aimed at the development of a multivariable based high-order fuzzy time series (FTS) forecast model that will be applied for forecasting handover success rate (mobility management) in the Lagos zone of Airtel Nigeria as a case study. Multivariable FTS is based on classifying data as primary (major) and secondary (minor) as against the conventional univariate approach adopted by most FTS approaches. In order to optimize the determination of an objective interval length, the kth fuzzy data clustering algorithm is adopted. In this work, 52 weeks data were collected from Airtel Lagos zone namely; handover success rate (HOSR), stand alone dedicated control channel (SDCCH), received signal strength (RSS), path loss (PLOSS) and bit error rate (BER). HOSR is termed the major data since it is the primary variable of interest whilst the rest are termed minor since they are secondary variables. 40 weeks of data are used as training data whilst 12 weeks are used as validation data. All the 40 weeks of data have been clustered in this work and the mid-value of each cluster is bounded and used in the fuzzification process (conversion of numeric values to fuzzy variables). The fuzzified data is then defuzzified and then used to forecast for week 41. The whole process is then repeated with the data length increasing each time by adding the training data of the current forecasted week to the existing training data, until forecasted results of week 41 to week 52 were obtained one after the other. The forecasted results obtained agreed with the actual data with a maximum variation of less than ±5%. The forecasting process of this type has a high computational cost in proportion to the data length and as such, a computer-based model was developed using a program written in MATLAB. The model was validated using the validation data of this work (from week 41 to week 52) and the results obtained from the developed computer-based model. To further enrich the model so as to serve as a verifiable basis of comparison and standard, established models (Chen, Mu’azu and Jilani) were applied to the same data set and results obtained were used as additional validation. The developed model has demonstrated that it can be used to forecast HOSR using SDCCH, RSS, PLOSS and BER as attributes due to its degree of consistency with respect to the result obtained from the statistical analysis carried out. The statistical values obtained are: Average Performance Error (APE) of 0.6615%, Maximum Performance Error (MPE) of 2.4841%, Pearson’s Correlation Coefficient (PCC) of 0.9800 and Root Mean Square Error (RMSE) of 0.05012. In addition, the developed model was compared with three other FTS models, the result differences are (APE (0.0397, 0.0207 and 0.0209 respectively), MPE (3.3122, 3.9251 and 3.0111 respectively), PCC (0.0291, 0.0099 and 0.0139 respectively) and RMSE (0.001542, 0.000989 and 0.00149 respectively)) all higher than those of the developed model. Therefore, the statistical analysis and validation with other FTS models have further ascertained and confirmed the accuracy of the developed model.