Forecasting sales is a critical aspect of business decision-making. As data continues to be at the heart of strategy planning, this blog discusses how we approached a univariate time series analysis project to predict sales for an electronics and appliance store.

Problem Statement

The goal of this project was to forecast the monthly sales for an electronics and appliance store for the next 12 months using past sales data.

Data Preparation

We began by loading the data, which consisted of two columns: Date and Sales. After ensuring data types were correctly assigned, we set the Date column as the index to establish a time series format,enabling us to analyze sales patterns over years.

Understanding Stationarity

Stationarity in time series analysis is key for accurate modeling. A stationary series has constant mean and variance over time. Using both visual representation and the Augmented Dickey-Fuller Test (ADF), we checked stationarity. Initially, the series was non-stationary, indicating a need for transformations like differencing to stabilize the mean and variance.

Transformation Techniques

We applied differencing techniques to achieve stationarity, followed by taking the logarithmic values of sales data. These transformations revealed underlying patterns, allowing us to decompose the series into its core components: trend, seasonality, and residuals.

Building the Time Series Forecast Model

The project explored several time series models like ARIMA, SARIMA and SARIMAX. PACF and ACF plots helped identify optimal parameters for these models. Using the auto_arima function, the SARIMAX model
was identified as the best fit. The SARIMAX model effectively captured the seasonal and trend components and was trained on the transformed data. The model was then used to forecast sales for the
next 12 months.

Conclusion

This univariate time series analysis demonstrated the power of data in forecasting future trends. The project not only provided accurate predictions but also showcased the importance of preparing and understanding data, testing stationarity, and selecting the right model. This process is a stepping-stone for businesses aiming to make data-driven decisions.