Banca de QUALIFICAÇÃO: ESDRAS ALEX FREIRE DE OLIVEIRA

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
STUDENT : ESDRAS ALEX FREIRE DE OLIVEIRA
DATE: 21/05/2024
TIME: 14:30
LOCAL: Plataforma Virtual Google Meet
TITLE:

Forecasting time series of products sold at retail

KEY WORDS:

Forecast. Time series. Stacked generalization. Machine learning. Product families.

PAGES: 97
BIG AREA: Ciências Agrárias
AREA: Agronomia
SUMMARY:

Forecasting the sales volume of retail product families is an indispensable step in the respective strategic business planning and efficient sizing of supply chain management. This practice allows organizations to anticipate demand for products, improving preparation to meet consumer needs. In this context, the adoption
of machine learning (ML) models for time series forecasting emerges as a valuable tool. In the present research, the objective was to analyze the predictive capacity of individual and combined mathematical models through a stacked generalization approach, to forecast time series of the sales volume of twenty product families. The individual models (or base models) were trained and, subsequently, the five best predictions among the k-NN, SVR, DTR, RFR, ETR, XGBoost, LightGBM, MLP and LSTM models were selected based on their performance on the MAAPE metric, due to its robustness in analyzing time series with high variability. In turn, the combined models (or meta-models) used the outputs of the five base models selected in the first stage to make the final prediction, and the four best predictions for each time series were subsequently selected from among the k-NN meta-models, SVR, DTR, MLP and LSTM, as well as the use of combiners based on descriptive statistics measures, such as the mean, median and minimum variance. The experiments involving the crossing between models resulted in 100 combinations between base models and 80 combinations between meta-models for the set of time series analyzed in this research. Subsequently, a performance comparison was carried out between the proposed models using the MAAPE, RMSLE and MAE metrics. A comparative analysis of the time series analyzed showed that, in general, the combined models stood out in relation to the individual models, presenting a marginally superior performance between the MAAPE, RMSLE and MAE metrics. These experiments contextualize the complexity of forecasting sales of product families and contribute to the literature regarding the optimization and applicability of ML models in predicting sales volume time series in companies in the retail segment.

COMMITTEE MEMBERS:
Externo à Instituição - BERNARDO SOBRINHO SIMÕES ALMADA-LOBO
Interno - ***.218.133-** - CÍCERO CARLOS FELIX DE OLIVEIRA - IFCE
Interno - FRANCISCO ALIXANDRE AVILA RODRIGUES
Presidente - PAULO RENATO ALVES FIRMINO