Time Series Methods

Time series models are used to forecast events based on verified historical data. Common types
include ARIMA, smooth-based, and moving average. Not all models will yield the same results
for the same dataset, so it's critical to determine which one works best based on the individual
time series.

When forecasting, it is important to understand your goal. To narrow down the specifics of
your predictive modeling problem, ask questions about:

1. Volume of data available — more data is often more helpful, offering greater
opportunity for exploratory data analysis, model testing and tuning, and model fidelity.
2. Required time horizon of predictions — shorter time horizons are often easier to
predict — with higher confidence — than longer ones.
3. Forecast update frequency — Forecasts might need to be updated frequently over
time or might need to be made once and remain static (updating forecasts as new
information becomes available often results in more accurate predictions).
4. Forecast temporal frequency — Often forecasts can be made at lower or higher
frequencies, which allows harnessing down-sampling and up-sampling of data (this in
turn can offer benefits while modeling).

ARIMA
The ARIMA model (an acronym for Auto-Regressive Integrated Moving Average), essentially
creates a linear equation which describes and forecasts your time series data. This equation is
generated through three separate parts which can be described as:

• AR — auto-regression: equation terms created based on past data points

• I — integration or differencing: accounting for overall “trend” in the data
• MA — moving average: equation terms of error or noise based on past data points

Together, these three parts make up the AR-I-MA model.

The AR and MA aspects of ARIMA actually come from standalone models that can describe
trends of more simplified time series data. With ARIMA modeling, you essentially have the
power to use a combination of these two models along with differencing (the “I”) to allow for
simple or complex time series analysis. Pretty cool, right?

ARIMA model is generally denoted as ARIMA(p, d, q) and parameter p, d, q are defined as

follow:

• p: the lag order or the number of time lag of autoregressive model AR(p)
• d: degree of differencing or the number of times the data have had subtracted with past
value
• q: the order of moving average model MA(q)

Measures of forecast accuracy

There are several measures to measure forecast accuracy:

• Mean Forecast Error (MFE)

 • Mean Absolute Error (MAE) or Mean Absolute Deviation (MAD)
• Root Mean Square Error (RMSE)
• Mean Absolute Percentage Error (MAPE)

Let us consider the following table for this example.

The table shows the weekly sales volume of a store. The store manager is not savvy with
forecasting models, so he considered the current week’s sales as a forecast for next week. This
methodology is also known as the naïve forecasting method due to the nature of simplicity.

Calculating Forecast Error

The difference between the actual value and the forecasted value is known as forecast error.
So, in this example, forecast error for week 2 is

Forecast Error (week 2) = 21 – 17 = 4

A positive value of forecast error signifies that the model has underestimated the actual value
of the period. A negative value of forecast error signifies that the model has overestimated the
actual value of the period.
The following table calculates the forecast error for the rest of the weeks:

Mean Forecast Error

A simple measure of forecast accuracy is the mean or average of the forecast error, also known
as Mean Forecast Error.

In this example, calculate the average of all the forecast errors to get mean forecast error:
The MFE for this forecasting method is 0.2.

Since the MFE is positive, it signifies that the model is under-forecasting; the actual value tends
to more than the forecast values. Because positive and negative forecast errors tend to offset
one another, the average forecast error is likely to be small. Hence, MFE is not a particularly
useful measure of forecast accuracy.

Mean Absolute Deviation (MAD) or Mean Absolute Error (MAE)

This method avoids the problem of positive and negative forecast errors. As the name suggests,
the mean absolute error is the average of the absolute values of the forecast errors.

MAD for this forecast model is 4.08

Mean Squared Error (MSE)
Mean Squared Error also avoids the challenge of positive and negative forecast errors offsetting
each other. It is obtained by:

• First, calculating the square of the forecast error

• Then, taking the average of the squared forecast error

Root Mean Squared Error (RMSE)

Root Mean Squared Error is the square root of Mean Squared Error (MSE). It is a useful metric
for calculating forecast accuracy.

RMSE for this forecast model is 4.57. It means, on average, the forecast values were 4.57
values away from the actual.
Mean Absolute Percentage Error (MAPE)
The size of MAE or RMSE depends upon the scale of the data. As a result, it is difficult to
make comparisons for a different time interval (such as comparing a method of forecasting
monthly sales to a method forecasting a weekly sales volume). In such cases, we use the mean
absolute percentage error (MAPE).
Steps for calculating MAPE:
• by dividing the absolute forecast error by the actual value
• calculating the average of individual absolute percentage error

The MAPE for this model is 21%.

It signifies that the 21% average deviation of the forecast from the actual value in the given
model.
STL approach
STL uses LOESS (locally estimated scatterplot smoothing) to extract smooths estimates of the
three components. The key inputs into STL are:

• season - The length of the seasonal smoother. Must be odd.

• trend - The length of the trend smoother, usually around 150% of season. Must be
odd and larger than season.
• low_pass - The length of the low-pass estimation window, usually the smallest odd
number larger than the periodicity of the data.

Forecasting with STL

STLForecast simplifies the process of using STL to remove seasonalities and then using a
standard time-series model to forecast the trend and cyclical components.

Here we use STL to handle the seasonality and then an ARIMA(1,1,0) to model the
deseasonalized data. The seasonal component is forecast from the find full cycle where

Extract features from generated model as Height, Average Energy etc and
Analyze for prediction

For Height Feature:

Global Mapper can automate the process of locating features on the ground by using Lidar
classified data with the automated feature extraction tools available with the Lidar Module.
The default parameter values can be changed by the user for improved results.

The Extract Vector Features tool can be used on classified ground points, either on an entire
point cloud or on a user-defined subsection. This lidar feature extraction tool lets the user derive
features such as building footprints, building roof structures, power lines, and other structures
from classified Lidar ground points.

Example:

The building extraction tool will create 3D building vectors or mesh features based on points
in the Structure/Building Point group.

The default display of extracted 3D buildings is controlled by Area Feature Types: Building -
Floor, Building - Ground, Building - Roof, and Building - Wall. These styles carry over to 3D
model buildings as well.

Simplified Building Footprints - Select this option to extract only the building footprints and
other similar features from the classified ground points.

Pin Footprints to Height - Use this option to set the elevation for all generated building
footprints features to a single value.
Simplified Planes - Select this option to identify and create roof features for extracted
buildings.

Simplified Side Walls - Select this to generate side wall features extracted buildings.

Sharpen edges and stitch planes by adding points at planar intersections - This option will
generate point features along intersections between planes. The points will be placed between
pairs of nearby points that are contained in different planes.

Buildings as Mesh - If selected, the building roof will each be turned into a single 3D mesh
feature.

Color Vertices by Lidar Intensity - Use this option to color the mesh vertices by the lidar
intensity value at that location.

Footprint Line Simplification Epsilon - Use this field to specify, in meters, the distance from
one vertex to another for the vertex to be removed when simplifying the building footprint
features. Larger values will further simplify the extracted building features.

Maximum Distance to Plane - Set the maximum distance in meters from an identified plane
a point can be to still be considered part of the plane.

Minimum Number of Points in Plane - Use this field to set the minimum number of points
needed in an area to identify a plane and create an extracted building feature.

Minimum Footprint Area - User this field to specify the minimum size of a building footprint
in square meters. Generated footprints smaller than the entered value will be discarded.

For Energy Feature:

To train any statistical or ML model, we need to first extract useful features from the data. For
example to Audio feature extraction is a necessary step in audio signal processing, which is a
subfield of signal processing. It deals with the processing or manipulation of audio signals. It
removes unwanted noise and balances the time-frequency ranges by converting digital and
analog signals. It focuses on computational methods for altering the sounds.

Different features capture different aspects of sound. Generally audio features are categorised
with regards to the following aspects:

• Level of Abstraction: High-level, mid-level and low-level features of musical signals.

• Temporal Scope: Time-domain features that could be instantaneous, segment-level
and global.
• Musical Aspect: Acoustic properties that include beat, rhythm, timbre (colour of
sound), pitch, harmony, melody, etc.
• Signal Domain: Features in the time domain, frequency domain or both.
• ML Approach: Hand-picked features for traditional ML modeling or automatic feature
extraction for deep learning modeling.

How do we categorize audio features at various levels of abstraction?

These broad categories cover mainly musical signals rather than audio in general:
• High-level: These are the abstract features that are understood and enjoyed by humans.
These include instrumentation, key, chords, melody, harmony, rhythm, genre, mood,
etc.
• Mid-level: These are features we can perceive. These include pitch, beat-related
descriptors, note onsets, fluctuation patterns, MFCCs, etc. We may say that these are
aggregation of low-level features.
• Low-level: These are statistical features that are extracted from the audio. These make
sense to the machine, but not to humans. Examples include amplitude envelope, energy,
spectral centroid, spectral flux, zero-crossing rate, etc.