This article is a continuation of the part one article " New Methods to Forecast Passenger Demand. Part 1"  We'll pick up right where we left off with the results of the processes suggested in part one.

The successful, yet tedious, implementation of the neural network to each of the five scenarios produced varying results.  Due to the tremendous effect September 11, 2001, had on U.S. domestic air travel, the resultant mean average error was calculated for the entire data population encompassing the years 2000 through Q3, 2005; however, the year including September 11, 2001 (Q3 2001 - Q2 2002), was extracted to identify how much prediction error was attributable specifically to that highly unpredictable period.  Even though the hypothesis was not fully supported by the data, this categorization of results should provide further insight as will be discussed later.

Table 2 - Mean Absolute Deviation (%)

ATL - LAX

Figure 1 - ATL - LAX Predicted and Actual Passengers Q1 2000 - Q3 2005

The predictions for the ATL - LAX market clearly showed the effects of September 11, 2001, as well as the effects of the entrance of low cost carriers into the market.  With JetBlue's introduction to the ATL - LGB market in Q2, 2003, the model did react in the appropriate direction; however, the large change in market dynamics produced an unstable condition in the following year.  Additional non-stop markets shown to affect the ATL - LAX market included:  ATL - LGB (Long Beach Airport/Daugherty Field), ATL - BUR (Burbank Bob Hope Airport), ATL - SNA (John Wayne - Orange County Airport), and ATL - ONT (Ontario International Airport).