EXECUTIVE SUMMARY   
   - A simple neural net (NN) is created to predict in advance California   
     monthly UFO sightings as per NUFORC database.  Some adjustments were   
     made to allow for the intro of a web report form in approx March 2006.   
   - The input data are planetary positions as per an ephemeris nominally   
     tuned to 2000-2050 but supposedly of reasonable accuracy back to   
     1950. Other data include surface temperature measures from ground   
     stations and satellites (the JMA gridded monthly dataset).   
   - The model is able to predict monthly California UFO sightings 12m in   
     advance to an accuracy of around +-2. In particular, the NN is   
     trained to predict "outliers" or UFO sighting clusters in preference   
     to the mundane seasonal month-to-month average counts. But it seems to   
     perform adequately for all months in the data.   
   - Similar models built to predict sightings in other US states were   
     not always as successful although many were. It seems some US states   
     generate sightings similar to "white noise". This does not mean   
     those data are "junk" and all California sightings are "something".   
     The model is not built to decide whether a particular predicted   
     sighting is the result of "weather", "alcoholic haze", or "unusual   
     physical object".   
   - Further work is in train to produce an annual or monthly UFO   
     prediction s/w that might be useful to pre-position or at least   
     anticipate UFO reports from certain regions up to 12m in advance.   
      
      
   This article is a outline of a very simple neural network that   
   predicts UFO sightings from sat-based surface temperature data and an   
   ephemeris of various planetary positions.   
      
   These data were not chosen because they are the best correlates of UFO   
   sightings data, but because they are reliably curated and updated by   
   govt agencies around the world and will provide a continuing quality   
   data stream into the future. The ephemeris data is generated by a   
   program distilled from tabulated data that can also be updated into   
   the future and also can be zoomed into hour by hour resolution if   
   necessary. Supposedly the Dec and RA for the planets in the dataset   
   are accurate to within a couple of arc minutes. Adequate for our   
   purposes here.   
      
   The s/w -- simulating a simple "brain" about the size of a sea squirt   
   (~200 neurons) -- takes a large set of monthly time series from the   
   online datasets and is trained to predict UFO sightings for California   
   from 1950-present as given by the NUFORC database. The present set-up   
   predicts sightings 12 months into the future. I.e. given data upto the   
   end of 2020 we can use the s/w to predict UFO activity in California   
   upto the end of 2021.   
      
   It turns out these sightings can be predicted to high precision given   
   they are only month by month resolution. The training procedure finds   
   the final error is within +-2 sightings for each month.  In particular   
   the error band applies to many of the "flaps" or sightings clusters   
   that have been seen in Cal from 1950.   
      
   We can not decide just here whether the "weather data" or the "planet   
   data" has a larger influence on the accuracy of the predictions.  In   
   any case, simply building a "good predictor" can't tell you who is   
   operating the UFO objects (or whatever) in question or whether some   
   large fraction of the sightings are e.g. weather-induced   
   hallucinations of some kind.   
      
   But it quite interesting the input data chosen can predict the   
   "outliers" so well.   
      
   Here's a sample of the "observed" versus "predicted" California   
   sightings since 2000:   
      
   Date of data	Adjusted(*) NUFORC sightings for California   
   		for 1 year later   
   YYYY.MM		OBS	PRED   
   2000.04		44	44   
   2000.12		55	55   
   2000.21		44	44   
   2000.29		55	55   
   2000.38		44	44   
   2000.46		110	110   
   2000.54		110	109	<- prediction too low by 1   
   2000.62		55	55   
   2000.71		131	131   
   2000.79		66	65	<- 1   
   2000.88		55	54	<- 2   
   2000.96		55	55   
   2001.04		55	54   
   2001.12		44	44   
   2001.21		44	44   
   2001.29		22	22   
   2001.38		55	56   
   2001.46		88	88   
   2001.54		44	43   
   2001.62		88	88   
   2001.71		55	55   
   2001.79		44	44   
   2001.88		55	55   
   2001.96		22	21   
   2002.04		44	43   
   2002.12		22	21   
   2002.21		22	22   
   2002.29		55	55   
   2002.38		44	44   
   2002.46		66	66   
   2002.54		88	88   
   2002.62		66	66   
   2002.71		44	44   
   2002.79		44	43   
   2002.88		55	55   
   2002.96		55	55   
   2003.04		55	56   
   2003.12		22	21   
   2003.21		22	22   
   2003.29		110	109	<-- predicts cluster   
   2003.38		66	66   
   2003.46		153	153	<-- predicts cluster   
   2003.54		66	66   
   2003.62		22	22   
   2003.71		88	88   
   2003.79		88	88   
   2003.88		22	22   
   2003.96		22	22   
   2004.04		44	44   
   2004.12		22	21   
   2004.21		44	45   
   2004.29		88	88   
   2004.38		44	44   
   2004.46		131	131	<-- predicts cluster   
   2004.54		153	153	<-- predicts cluster   
   2004.62		66	66   
   2004.71		22	22   
   2004.79		66	65   
   2004.88		66	65   
   2004.96		55	55   
   2005.04		44	44   
   2005.12		131	131	<-- predicts cluster   
   2005.21		11	11   
   2005.29		33	33   
   2005.38		36	36   
   2005.46		34	34   
   2005.54		62	62   
   2005.62		52	52   
   2005.71		44	44   
   2005.79		44	44   
   2005.88		54	54   
   2005.96		66	66   
   2006.04		55	55   
   2006.12		31	31   
   2006.21		59	59   
   2006.29		36	36   
   2006.38		30	30   
   2006.46		46	47   
   2006.54		57	57   
   2006.62		50	50   
   2006.71		42	42   
   2006.79		48	48   
   2006.88		70	70   
   2006.96		58	58   
   2007.04		72	71   
   2007.12		52	52   
   2007.21		49	49   
   2007.29		59	59   
   2007.38		44	44   
   2007.46		65	65   
   2007.54		38	38   
   2007.62		54	54   
   2007.71		57	57   
   2007.79		77	77   
   2007.88		83	83   
   2007.96		50	50   
   2008.04		98	97   
   2008.12		61	60   
   2008.21		47	47   
   2008.29		30	30   
   2008.38		34	34   
   2008.46		55	55   
   2008.54		65	65   
   2008.62		66	66   
   2008.71		54	54   
   2008.79		42	42   
   2008.88		43	44   
   2008.96		44	44   
   2009.04		52	52   
   2009.12		30	30   
   2009.21		33	33   
   2009.29		38	38   
   2009.38		42	43   
   2009.46		42	42   
   2009.54		78	78   
   2009.62		63	63   
   2009.71		67	67   
   2009.79		52	51   
   2009.88		57	57   
   2009.96		41	41   
   2010.04		46	46   
   2010.12		54	53   
   2010.21		42	42   
   2010.29		52	51   
   2010.38		39	38   
   2010.46		37	37   
   2010.54		60	60   
   2010.62		70	70   
   2010.71		47	47   
   2010.79		55	55   
   2010.88		50	50   
   2010.96		65	66   
   2011.04		63	63   
   2011.12		35	34   
   2011.21		48	48   
   2011.29		57	57   
   2011.38		51	51   
   2011.46		79	79   
   2011.54		69	69   
   2011.62		75	75   
   2011.71		70	70   
   2011.79		73	73   
   2011.88		71	71   
   2011.96		60	59   
   2012.04		40	40   
   2012.12		45	45   
   2012.21		52	52   
   2012.29		53	53   
   2012.38		56	56   
   2012.46		58	58   
   2012.54		67	67   
   2012.62		64	64   
   2012.71		61	61   
   2012.79		57	57   
   2012.88		64	64   
   2012.96		102	102	<-- predicts cluster   
   2013.04		90	89   
   2013.12		73	72   
   2013.21		61	61   
   2013.29		77	77   
   2013.38		60	60   
   2013.46		80	80   
   2013.54		72	72   
   2013.62		79	79   
   2013.71		66	66   
   2013.79		68	67   
   2013.88		55	55   
   2013.96		67	67   
   2014.04		60	60   
   2014.12		39	40   
   2014.21		45	46   
   2014.29		34	34   
   2014.38		36	36   
   2014.46		34	34   
   2014.54		53	53   
   2014.62		48	48   
   2014.71		73	74   
   2014.79		72	73   
   2014.88		250	251	<-- predicts cluster   
   2014.96		50	51   
   2015.04		38	39   
   2015.12		64	65   
   2015.21		38	39   
   2015.29		34	34   
   2015.38		23	23   
   2015.46		49	50   
   2015.54		91	91   
   2015.62		51	51   
   2015.71		63	64   
   2015.79		40	41   
   2015.88		42	42   
   2015.96		36	36   
   2016.04		31	31   
   2016.12		34	34   
   2016.21		33	33   
   2016.29		51	51   
   2016.38		47	47   
   2016.46		37	37   
   2016.54		48	48   
   2016.62		46	46   
   2016.71		48	48   
   2016.79		58	58   
   2016.88		46	47   
   2016.96		95	95   
   2017.04		34	34   
   2017.12		42	42   
   2017.21		20	21   
   2017.29		16	16   
   2017.38		23	23   
   2017.46		10	11   
   2017.54		36	36   
   2017.62		32	32   
   2017.71		25	25   
      
   [continued in next message]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)