This document discusses time series forecasting of gas consumption data from January 1956 to August 1995. It begins with an exploratory data analysis of the data, examining trends, seasonality, and outliers. Both additive and multiplicative models are considered. The data is found to be non-stationary. The data is then split into training and test sets. Several forecasting models are applied including simple, exponential, Holt's linear trend, and Holt-Winters seasonal models. The accuracy of each model is evaluated on the test set using various error metrics. The best performing models are able to capture the trend and seasonality in the gas consumption time series.
This document discusses time series forecasting of gas consumption data from January 1956 to August 1995. It begins with an exploratory data analysis of the data, examining trends, seasonality, and outliers. Both additive and multiplicative models are considered. The data is found to be non-stationary. The data is then split into training and test sets. Several forecasting models are applied including simple, exponential, Holt's linear trend, and Holt-Winters seasonal models. The accuracy of each model is evaluated on the test set using various error metrics. The best performing models are able to capture the trend and seasonality in the gas consumption time series.
This document discusses time series forecasting of gas consumption data from January 1956 to August 1995. It begins with an exploratory data analysis of the data, examining trends, seasonality, and outliers. Both additive and multiplicative models are considered. The data is found to be non-stationary. The data is then split into training and test sets. Several forecasting models are applied including simple, exponential, Holt's linear trend, and Holt-Winters seasonal models. The accuracy of each model is evaluated on the test set using various error metrics. The best performing models are able to capture the trend and seasonality in the gas consumption time series.
This document discusses time series forecasting of gas consumption data from January 1956 to August 1995. It begins with an exploratory data analysis of the data, examining trends, seasonality, and outliers. Both additive and multiplicative models are considered. The data is found to be non-stationary. The data is then split into training and test sets. Several forecasting models are applied including simple, exponential, Holt's linear trend, and Holt-Winters seasonal models. The accuracy of each model is evaluated on the test set using various error metrics. The best performing models are able to capture the trend and seasonality in the gas consumption time series.
Download as DOCX, PDF, TXT or read online from Scribd
Download as docx, pdf, or txt
You are on page 1/ 17
Time Series
Forecasting- Gas
Dr. Areej Aftab Siddiqui Siddiqui
PGP-BABI (June 2019)
0 Contents 1. PROJECT OBJECTIVE.......................................................................................................................2 2. EXPLORATORY DATA ANALYSIS – STEP BY STEP APPROACH..........................................................2 3. Exploratory Data Analysis..............................................................................................................2 3.1 Analysis..................................................................................................................................2 3.2 Additive and Multiplicative model.........................................................................................4 4. Stationarity....................................................................................................................................7 5. Train and Test................................................................................................................................8 6. Forecasting....................................................................................................................................8 6.1 Simple Forecasting.....................................................................................................................8 6.2 Exponential Forecasting on Gas..................................................................................................9 6.3 Holts method...............................................................................................................................9 6.4 Removing seasonality from the time series and testing Holt's efficacy again...........................10 6.6 Holt winters...............................................................................................................................11 6.7 Holt winters multiplicative model..............................................................................................12 7. ARIMA and Auto ARIMA..............................................................................................................13 8. Model Accuracy...........................................................................................................................15 8.1 Ljung box test............................................................................................................................15 8.2 Box-Pierce test...........................................................................................................................15 8.3 Accuracy of the forecast............................................................................................................15
1 1. PROJECT OBJECTIVE
The objective of the report is to explore the datasets on the Gas and to perform Principal Component Analysis and Regression on the variables. The Case Study will be solved using R programming. The report will mainly consist of the following: i. Importing the dataset in R ii. Examining the structure of the Gas dataset iii. Graphically depicting the Gas dataset iv. Examining the presence of stationarity v. Forecast for 20 periods vi. Check accuracy
2. EXPLORATORY DATA ANALYSIS – STEP BY STEP APPROACH
A Typical Data exploration activity consists of the following steps: 1. Environment Set up and Data Import 2. Variable Identification 3. Analysis 4. Missing Value Treatment 5. Forecasting 6. Accuracy Check
We shall follow these steps in exploring the provided dataset.
Environment Set Up and Data Import
Install necessary Packages and Invoke Libraries The packages installed and libraries invoked for the current project are tidyverse, Metrics, dplyr, vif, car, corrplot Set up working Directory The working directory is set to import and export data files. For the current project the working directory was set in Documents in C Drive. Import and Read the Dataset The given dataset is in .xlsx format. Hence, the command ‘read.xlsx is used for importing the file.
Please refer Appendix A for Source Code.
3. Exploratory Data Analysis
3.1 Analysis The frequency of the dataset is 12 (indicating months). Table-1 Summary Min. 1st Qu. Median Mean 3rd Qu. Max. 1646 2287 3345 9425 15996 43558
2 There is no missing value. The dataset starts from January 1956 and ends on August 1995. Summary output of the data indicates that the dataset might be skewed. The Monthly, Quarterly and Annual Gas trends indicate that Gas consumption rises after 1980 with high rate of growth after 1985.
On observing the Seasonality, it is seen that Gas consumption rises from April, is maximum in July and then decreases.
The dataset has outliers in the month of March, May, July and August.
3 Boxplot
It is important to adjust the time series as each month has different number of days. The adjusted series trend is as follows:
3.2 Additive and Multiplicative model
In an additive model, the amplitude of both the seasonal and irregular variations do not vary as the level of the trend rises or falls. Decomposition is a tool that can separate different components in time series. The decomposition of additive series indicates, rising trend with similar seasonality each year.
4 Similar trend is observed in case of Multiplicative time series
5 Loess ("locally-weighted scatterplot smoothing") uses local regression. A window of a specified width is placed over the data. The wider the window, the smoother the resulting loess curve. If you think the seasonal pattern is constant through time, you should set this parameter to a big value, so that you use the entire data to perform your analysis.
6 Summary: Seasonality is strong but consistent. Trend - 1950 to 1970, there was a strong growth and after that, the noises went up started from 1990 as well. Additive and Multiplicative gives same results. Hence, the dataset (time series) is additive. As the time series increases in magnitude, the seasonal variation increases as well.
4. Stationarity
Stationarity is important in time series forecasting. It is seen through KPSS and ADF that time series is not stationary.
data: gas.ts Dickey-Fuller = 0.73972, Lag order = 6, p-value = 0.99 alternative hypothesis: stationary
7 ACF and PACF plots ( another method to observe stationarity)
It is seen that the dataset is not stationary.
5. Train and Test
Train data is taken from 1970 January to 1993 December and test data for 1994 only. Simple moving average is used to visualise time series.
6. Forecasting 6.1 Simple Forecasting On plotting the forecast it is seen that for 20 periods,
For 12 periods
8 Accuracy: ME RMSE MAE MPE MAPE ACF1 Theil's U Test set 2153.958 4904.692 3809.181 5.203979 12.32112 0.703002 1.638023 s The forecasts indicate rising trend in the next 12 periods.
6.2 Exponential Forecasting on Gas
Smoothing parameters: alpha = 0.2
Initial states: l = 1928.6188
sigma: 1702.584
AIC AICc BIC
5918.310 5918.352 5925.635
ME RMSE MAE MPE MAPE MASE ACF1
Training set 395.33 1696.662 1061.25 1.618 14.300 1.079 0.804 Test set 4131.69 6400.524 5097.60 11.789 16.201 5.18 0.7365
9 Theil's U Training set NA Test set 2.114935 The forecasts indicate stagnant trend in the next 12 periods.
5587.001 5587.214 5605.316 ME RMSE MAE MPE MAPE Training set -8.404 944.648 575.2032 0.4822 7.538 Test set 17680.00 19818.83 17680.000 58.4034 58.40 MASE ACF1 Theil's U Training set 0.5849 -0.00282 NA Test set 17.979 0.777285 6.928428 The forecasts indicate falling trend in the next 12 periods.
10 6.4 Removing seasonality from the time series and testing Holt's efficacy again
ME RMSE MAE MPE MAPE MASE
Training set -6.033996 755.8903 558.1407 4.162145 16.48117 0.5676028 Test set 11335.864070 12711.9230 11335.8641 37.619269 37.61927 11.5280398 ACF1 Theil's U Training set 0.03661692 NA Test set 0.77404723 6.231689 The forecasts indicate falling trend in the next 12 periods after removing seasonality.
Initial states: l = 2230.6766 b = 7.2177 s = -1034.848 -532.6302 71.5714 516.9939 1504.973 1781.418 1320.311 588.8141 -554.9172 -747.7943 -1420.407 -1493.485
sigma: 605.0447
AIC AICc BIC
5338.864 5341.407 5404.798
Training set error measures
ME RMSE MAE MPE MAPE MASE ACF1 Training set 81.59 586.91 387.415 0.9087 7.175 0.3939 0.4235
Jan Feb Mar Apr May Jun Jul Aug
1994 21010.61 21398.96 23621.92 25059.73 27968.26 29980.18 33316.35 32392.95 Sep Oct Nov Dec 1994 28465.48 27323.58 25045.88 23600.47
accuracy(gashw.f1, gas.test) ME RMSE MAE MPE MAPE MASE Training set 81.59189 586.9159 387.415 0.9087759 7.175919 0.3939828 Test set 2233.13630 2602.5705 2233.136 7.2892297 7.289230 2.2709944 ACF1 Theil's U Training set 0.4235979 NA Test set 0.4241718 0.9189364 The forecasts indicate seasonal trend in the next 12 periods.
Initial states: l = 2034.7689 b = 3.6615 s = 0.8541 0.917 1.0114 1.0758 1.218 1.2715 1.1749 1.0929 0.9184 0.8838 0.7932 0.789
sigma: 0.0454
AIC AICc BIC
4773.207 4775.750 4839.140
Training set error measures:
ME RMSE MAE MPE MAPE MASE Training set 27.94532 434.7321 253.4229 0.3438495 3.183346 0.2577191 ACF1 Training set -0.07376761 ME RMSE MAE MPE MAPE MASE Training set 27.94532 434.7321 253.4229 0.3438495 3.183346 0.2577191 Test set 2197.42260 2341.3448 2197.4226 7.5092231 7.509223 2.2346753 ACF1 Theil's U Training set -0.07376761 NA Test set 0.07767108 0.8533396 The forecasts indicate seasonal trend in the next 12 periods. Holt Winters multiplicative model is most accurate considering MAPE, MAE and RMSE.
f7=forecast(gasarima) > accuracy(f7, gas.test) ME RMSE MAE MPE MAPE MASE Training set 33.71582 852.161 521.0072 0.4181353 6.813444 0.5298398 Test set 8224.60366 9645.597 8224.6037 26.3034427 26.303443 8.3640345 ACF1 Theil's U Training set -0.006466775 NA Test set 0.733738029 3.308731 f8=forecast(fit) > accuracy(f8, gas.test) ME RMSE MAE MPE MAPE MASE Training set 35.20191 852.7607 522.7044 0.4097849 6.844272 0.5315658 Test set 8036.47583 9464.9462 8036.4758 25.6577831 25.657783 8.1727174 ACF1 Theil's U Training set -0.02372956 NA Test set 0.73383680 3.24253
The ARIMA model forecasts gas consumption accurately.
