Project 07: Machine Learning and Planetary Cycles #

Source: bigastrologybook.com/2/research/19/project-7 Archive Date: 2026-03-21 Book: The Big Astrology Book of Research by Renay Oshop Datasets: OpenPsychometrics Big Five (N=19,632) + Verified Celebrity Charts (N=82, from Project 06)

Research Questions #

Two distinct questions are tested here under the shared umbrella of "machine learning and astrology":

1. Cyclic Age Hypothesis: Do planetary return cycles (Saturn at 29.5 years, Jupiter at 11.9 years, Lunar Nodes at 18.6 years) predict Big Five personality traits better than simple linear aging?
2. Profession Classification: Can a birth chart predict professional category (Science, Arts, Politics, Sports) better than chance?

Hypothesis #

If Saturn Returns and other cyclic life events shape personality meaningfully, then a model using cyclical phase position should outperform one using linear age. For profession classification, if planetary patterns carry information about career path, a machine learning classifier should exceed random guessing.

Why Two Experiments? #

The Big Five dataset provides personality data with only birth year, not full birth date. This prevents calculation of Sun, Moon, Mercury, Venus, or Mars positions — all of which cycle faster than a year. Only the outer planets (Jupiter through Pluto), whose periods span years to decades, can be extracted from annual data. This dataset is appropriate for testing cyclic-age hypotheses but not for full-chart classification.

The celebrity dataset (N=82 from Project 06) provides verified birth dates and times with known professional categories — making it appropriate for the second question.

Data Sources #

Experiment 1: The Cyclic Age Hypothesis #

Method #

From each participant's birth year, the analysis derived:

• Saturn cycle phase (0–1, period 29.5 years): where in the Saturn cycle the person was at time of survey
• Jupiter cycle phase (0–1, period 11.9 years)
• Nodal cycle phase (0–1, period 18.6 years)

These were added as features to a regression model predicting each Big Five trait. The key comparison: does adding cyclic phase improve prediction over simple linear age?

Inner planet aliasing note: Sun, Moon, Mercury, Venus, and Mars complete multiple cycles per year. Annual-resolution data cannot distinguish where in these fast cycles a person falls — a methodological constraint correctly acknowledged. Only the outer planets were testable.

Results: Cyclic Age vs. Linear Age #

Interpretation: Adding cyclic planetary phases over linear age provides effectively zero additional predictive power for any Big Five trait. Personality development tracks biological aging, not planetary return cycles. The "Saturn Return" (ages 29 and 58) shows no statistical spike in Neuroticism or Conscientiousness compared to adjacent ages.

The personality R² values themselves (0.02–0.05) are small but typical for age-personality research — age predicts some variance in maturity-related traits like Conscientiousness, but most personality variance is explained by factors other than age.

Multiple testing note: Five traits were tested. A Bonferroni-corrected threshold at α = 0.01 applies. No trait crossed significance in either the linear or cyclic model.

The Saturn Opposition Dip #

While the global model comparison showed no improvement from cyclic features, closer inspection of Conscientiousness residuals revealed a localized pattern at the Saturn Opposition (approximately ages 14, 44, and 73 — the midpoint of the cycle, not the Return).

This "Mid-Cycle Crisis" pattern is unconventional: traditional astrology emphasizes the Saturn Return (ages 29/58) as the key transition, not the Opposition. Yet the data suggests the midpoint dip may be behaviorally meaningful — perhaps corresponding to the adolescent and midlife periods where discipline and self-regulation are most under stress. The effect is small (~0.59 points on a 10–50 scale) and requires a formal effect size and p-value calculation before it can be treated as more than exploratory.

Experiment 2: Profession Classification #

Method #

Using the verified celebrity dataset (N=82), a Random Forest classifier was trained on full birth chart features:

• Planetary sign placements (Sun through Pluto), expressed as sin/cos components
• Element counts (Fire, Earth, Air, Water)
• Mode counts (Cardinal, Fixed, Mutable)

Target: Profession category (Science, Arts, Politics, Sports, and other categories). Validation: Leave-One-Out Cross Validation (LOO-CV), which tests each chart against a model trained on all others — appropriate for small N.

Results #

The classifier achieved 29.3% accuracy against a random baseline of roughly 16.7%. This is statistically above chance — correct 75% more often than guessing — but wrong 70.7% of the time.

Top Predictive Features #

Feature importance rankings from the Random Forest:

Critical caveat: The top two features (Pluto and Neptune signs) are essentially age proxies. Scientists born in 1940 share the same Pluto sign but also share the same cultural and institutional context. The classifier may be detecting historical patterns in who became famous in which category, not anything intrinsic to the chart. Mars placement, ranking third, is not confounded by generation in the same way and aligns with Project 06's finding that Mars carries real signal in astrological data.

Statistical Caveats #

• Sample size (Experiment 2): N=82 for LOO-CV produces wide confidence intervals (approximately ±10 percentage points at 95%). The 29.3% result is above chance but not robust enough to claim strong generalization.
• Generational confound: Outer planet sign placements (especially Pluto and Neptune) are strongly correlated with birth year (Spearman r ≈ 0.9). Any apparent predictive power from these features likely reflects historical era, not astrology.
• Annual resolution (Experiment 1): Inner planet cycles cannot be tested on birth-year-only data. The null result for cyclic age applies only to the outer planets.
• Multiple testing (Experiment 1): Five Big Five traits × two model types = 10 comparisons. No corrections were needed because no comparison approached significance.

Conclusion #

Experiment 1 (null): Cyclic planetary phases add nothing to the prediction of Big Five personality traits beyond simple linear age. The Saturn Return has no detectable signature in this dataset of 19,632 people. A marginal Saturn Opposition dip in Conscientiousness warrants further investigation but is not yet confirmatory.

Experiment 2 (modest above-chance): A Random Forest on full birth charts achieves 29.3% profession classification accuracy (vs. ~16.7% random), but this likely reflects generational/historical confounds in outer planet placements rather than genuine astrological signal. The Mars feature — not confounded by generation — provides the most interesting lead, consistent with Project 06's harmonic findings.

The methodological lesson is practical: annual-resolution data is insufficient to test fast-planet hypotheses. Future work should use full birth dates and times, and should partial out the generational confound by testing within narrow birth-year cohorts.

Archived code and raw data outputs preserved in backup/.

Machine Learning and Planetary Cycles #

Research Question #

"Can we predict personality traits using Planetary Cycles (Age) instead of static Birth Charts?"

Traditional astrology claims that life follows cyclic patterns (e.g., the "Saturn Return" at age 29.5). Critics argue that any "astrological" effect is simply Linear Aging (getting older makes you more conscientious).

In this project, we rigorously compare two models:

1. Linear Age Model: Traits change linearly with biological age.
2. Cyclic Age Model: Traits fluctuate according to planetary periods (Saturn=29.5y, Jupiter=11.9y, Nodes=18.6y).
3. Inner Planet Cycles: We also tested Sun, Moon, Mercury, Venus, and Mars, but confirmed that on annual-resolution data, these fast cycles are mathematically aliased and provide no valid signal.

Part B: Profession Classification (Inner Planets) #

Since the Big Five dataset lacked birth dates (preventing Inner Planet calculation), we utilized the Verified Celebrity Dataset (N=82) from Project 6 to perform a "Proof of Concept" ML Classification using the full chart (Sun through Pluto).

• Task: Predict Profession (Science, Arts, Politics, Sports, etc) from Chart Features.
• Features: Planetary Sign placements (Sin/Cos components), Element Counts, Mode Counts.
• Model: Random Forest Classifier (Leave-One-Out CV).

Execution #

Run the analysis:

python3 analysis_cyclic.py        # Age/Cycle Analysis
python3 analysis_ml_profession.py # Profession Classification

Data Provenance #

Personality Data #

• Source: OpenPsychometrics.
• Dataset: Big Five Personality Test (IPIP-NEO-300).
• Link: https://openpsychometrics.org/_rawdata/
• File: BIG5.zip.