TIME SERIES ANALYSIS

PATTERNS, FORECASTS, AND ANALYSIS

Prasad Deshmukh
 TIME SERIES ANALYSIS
• Time series analysis is a branch of data
analysis that focuses on studying and
understanding data points collected over
time.
• It deals with data that is recorded at regular
intervals, such as hourly, daily, monthly, or
yearly, and aims to uncover patterns, trends,
and relationships within the data.
• Time series analysis is used in various fields,
including finance, economics, weather
Prasad Deshmukh
forecasting, and stock market prediction.
 1. TIME SERIES DATA
Time Stamps: Temporal Order:
• Each observation in a time series data set is • The observations in time series data are
associated with a specific time stamp or a arranged in a chronological order, with earlier
point in time. observations preceding later ones.
• These time stamps can be represented in • The temporal order is essential for analyzing
different formats, such as dates, patterns and making predictions based on the
timestamps, or time intervals. historical sequence.
Time Intervals: Sequential Dependence:
• Time series data can be collected at • Time series data exhibits a sequential
regular intervals (e.g., hourly, daily, dependence, meaning that the value of an
monthly) or irregular intervals (e.g., observation is often influenced by previous
sporadic events, event-driven data). observations in the sequence.
• The frequency of data collection • This dependence can manifest as trends,
determines the level of granularity in seasonality, or other patterns in the data.
the time series. Continue….
Prasad Deshmukh
 1. TIME SERIES DATA
Anomaly Detection: Longitudinal Studies:
• Time series data can be analyzed to detect • Time series data is often used in longitudinal
anomalies or unusual events that deviate studies, where observations are collected
from the expected patterns. repeatedly from the same subjects or entities
• Anomalies can indicate abnormalities, over time.
errors, or significant events that require • This enables the study of changes and trends
attention or further investigation. in variables of interest within the subjects or
entities.
Time-based Analysis:
• Time series data analysis focuses on Time Series Models:
studying the behavior, patterns, and Autoregressive Integrated Moving Average
relationships within the data over time. (ARIMA), Seasonal ARIMA (SARIMA),
• It involves identifying trends, Exponential Smoothing (ETS),
seasonality, cyclical patterns, and other Recurrent Neural Networks (RNNs) and
temporal characteristics to gain insights Long Short-Term Memory (LSTM)
and make predictions. networks.
Prasad Deshmukh
 2. TREND ANALYSIS
Trend: Visual Inspection:
• A trend refers to the general direction or • Trend analysis often starts with visual
tendency of the data as it evolves over inspection of the time series plot to observe
time. any apparent long-term patterns or
• It represents the underlying movement or movements.
pattern that is not affected by short-term • Trend can be identified by looking for a
fluctuations or noise. consistent upward or downward movement
or any consistent oscillation.
Trend Types:
• Upward Trend: The data shows a consistent Mathematical Models:
increase or growth over time. • Various mathematical models can be used
• Downward Trend: The data shows a to estimate or quantify the trend
consistent decrease or decline over time. component of the data.
• Flat Trend: The data remains relatively stable • Examples include linear regression,
or exhibits no significant change over time. polynomial regression, or exponential
growth/decay models. Continue….
Prasad Deshmukh
 2. TREND ANALYSIS
Trend Estimation: Trend Evaluation:
• Trend estimation involves estimating the • Trend analysis also includes evaluating the
parameters of the trend model, such as significance and reliability of the identified
slope or growth rate. trend.
• This estimation helps to quantify and • Statistical tests or measures like p-values,
measure the strength or magnitude of the confidence intervals, or goodness-of-fit metrics
trend. can be used for this purpose.
Trend Breakpoints: Business Applications:
• Trend analysis can detect changes or • Trend analysis is widely used in various fields
breakpoints in the trend, indicating a shift such as finance, economics, sales forecasting,
or transition in the long-term pattern. demand planning, and market analysis.
• These breakpoints can signify structural • It helps in understanding historical patterns,
changes in the data or external factors making predictions, and supporting decision-
impacting the trend. making processes.
Prasad Deshmukh
 3. SEASONALITY ANALYSIS
Seasonality: Seasonal Patterns:
• Seasonality refers to the regular and • Seasonal patterns can take different forms,
predictable patterns that repeat over time such as a consistent increase or decrease in
within a time series. values during a particular season, peaks and
• These patterns can be daily, weekly, troughs at specific times, or periodic
monthly, quarterly, or annual cycles. fluctuations with a fixed duration.
Seasonal Adjustment: Seasonal Indices:
• Seasonal adjustment is the process of • Seasonal indices are factors used to
removing the seasonal component from the measure the relative strength or impact of
time series data to better understand and seasonality at different time points.
analyze the underlying trend and other • They provide information about the
components. magnitude of the seasonal effect
• This helps in identifying the non-seasonal compared to the average or trend in the
patterns and relationships. data.
Prasad Deshmukh Continue….
 3. SEASONALITY ANALYSIS
Seasonal Patterns and Decision Making:
Decomposition of Seasonality:
• Understanding and anticipating seasonal
• Seasonality can be decomposed into its
patterns can be valuable for businesses in terms
constituent components, which include the
of inventory management, resource allocation,
trend, seasonal, and residual components.
marketing campaigns, and sales forecasting.
• This decomposition helps in isolating and
• It allows organizations to make informed
analyzing the seasonal patterns separately.
decisions and optimize their operations
Multiple Seasonalities: accordingly.
• In some time series, there may be
Seasonality Testing:
multiple seasonal patterns occurring
• Statistical tests such as the Seasonal
simultaneously.
Decomposition of Time Series (STL) or
• For example, retail sales data may exhibit
Fourier analysis can be used to detect the
both a weekly and an annual seasonality.
presence and strength of seasonality in the
Analyzing and modeling multiple
data.
seasonalities can provide a more
• These tests help in validating the significance
comprehensive understanding of the data.
Prasad Deshmukh
of the identified seasonal patterns.
 4. SMOOTHING TECHNIQUES
Smoothing Techniques: Exponential Smoothing:
• Smoothing techniques in time • Exponential smoothing assigns exponentially
series analysis are used to reduce decreasing weights to the past observations, with
noise or irregularities in the data more weight given to recent observations.
and highlight the underlying • The most common types of exponential smoothing
patterns, trends, or components. are single exponential smoothing (SES), double
exponential smoothing (Holt's method), and triple
Moving Averages: exponential smoothing (Holt-Winters method).
• Moving averages smooth the time
series data by calculating the average Filtering Techniques:
of a fixed window or interval of • Filtering techniques such as Butterworth filters,
observations. Gaussian filters, or median filters can be used to
• Different types of moving averages smooth the time series data.
include simple moving average (SMA), • These techniques apply mathematical operations
weighted moving average (WMA), and to the data to reduce noise or high-frequency
exponential moving average (EMA). components.
Prasad Deshmukh
 5. STATIONARITY
Stationarity: Weak Stationarity:
• A stationary time series is one where • Weak stationarity, also known as second-order
the statistical properties such as stationarity or covariance stationarity, is a less
mean, variance, and covariance stringent condition. For a time series to be
remain constant or do not exhibit any weakly stationary, the
systematic change over different time • mean, variance, and autocovariance structure
periods. must be constant over time.
Strict Stationarity: Importance of Stationarity:
• A time series is said to be strictly • Stationarity is an important assumption in many
stationary if the joint probability time series models, as it simplifies the analysis
distribution of any set of observations is and facilitates accurate forecasting.
invariant to shifts in time. • Stationary time series are predictable, as their
• In other words, the statistical properties statistical properties remain constant, allowing
of the series, such as mean, variance, and for the use of statistical models that assume
covariance, do not change over time. stationarity.
Prasad Deshmukh
 6. FORECASTING
Time Series Models: Training and Testing:
• Time series models are mathematical models • To assess the accuracy and performance of
that capture the patterns and dependencies in a forecasting model, it is common practice
the time series data to make forecasts. to split the time series data into training and
• Common time series models include testing sets.
Autoregressive Integrated Moving Average • The model is trained on the historical data
(ARIMA), Seasonal ARIMA (SARIMA), and then tested on unseen future data to
Exponential Smoothing (ETS), and state- evaluate its forecasting ability.
space models.
Forecast Updating:
Forecast Evaluation: • As new data becomes available, forecasting
• Forecast accuracy can be evaluated using models can be updated to incorporate the
various metrics, such as mean absolute error most recent observations.
(MAE), mean squared error (MSE), or root • This allows for adaptive and up-to-date
mean squared error (RMSE). forecasts, particularly in dynamic and
changing environments. Prasad Deshmukh
In conclusion, time series analysis is a powerful framework that
allows us to understand and extract valuable insights from temporal
data. By examining trends, seasonality, and other patterns, we can
uncover underlying relationships and make informed forecasts.
Through techniques like smoothing, forecasting, and stationarity
analysis, time series analysis provides a robust toolkit for exploring
and interpreting data in various domains, supporting decision-
making and planning for the future.

Prasad Deshmukh
Prasad Deshmukh