DS 786 – Week 3
- Watch Videos
- Forecasting- Linear Trend
- Let’s consider more patterns
- Smoothing filters work best for stable timeseries data when there aren’t trends or seasonality
- Seasonality (like sine waves) is its own thing with different analytical techniques
- Trends up and down are not good for smoothing filters; regressions are
- Nonlinear trend analysis
- Not what we’re doing in this class
- Linear trend analysis
- y=mx+b
- Finding the intercept and slope lets us make predictions about the future
- All forecasts work better in the short-term than the long-term
- Associative Forecasting
- Maybe other variables are better for regressions than time.
- An associative forecast compares variables rather than using time as a variable.
- Linear regression
- Linear trend analysis
- y = intercept + slope * time
- Associative forecast
- y = intercept + slope * value of second variable at each time
- What makes regressions accurate
- r^2 is the percentage of variation in y that is explained by x
- correlation (r)
- Correlations can be positive and negative
- Positive: trend together
- Negative: trend in opposite directions
- Correlations can be spurious
- Correlation is signed, where r^2 gives an absolute value.
- Quiz