SMCI Stock Predictor Project

SMCI Stock Predictor Project#

Author: Haoyu Zhang
Course Project, UC Irvine, Math 10, Spring 25
I would like to post my notebook on the course’s website. [Yes]

I) Introduction#

The purpose of this project is to predict the stock price of SMCI (Super Micro Computer, Inc.). SMCI has shown significant growth and volatility in recent years, which makes it an interesting candidate for time-series modeling. We will explore its historical price data, visualize patterns, engineer features, and use regression models (linear and tree-based) to predict the next-day closing price.

II) Importing Data#

We import the required libraries and load SMCI’s historical stock data from a manually provided CSV file (MAX range).

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.linear_model import LinearRegression
from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error, r2_score

df = pd.read_csv("HistoricalData_1749502320765.csv")
df['Close/Last'] = df['Close/Last'].str.replace('$', '', regex=False).astype(float)
df['Open'] = df['Open'].str.replace('$', '', regex=False).astype(float)
df['High'] = df['High'].str.replace('$', '', regex=False).astype(float)
df['Low'] = df['Low'].str.replace('$', '', regex=False).astype(float)
df['Volume'] = df['Volume'].astype(str).str.replace(',', '').astype(int)
df['Date'] = pd.to_datetime(df['Date'])
df.sort_values('Date', inplace=True)
df.head()
Date Close/Last Volume Open High Low
2514 2015-06-09 3.344 2886240 3.363 3.379 3.321
2513 2015-06-10 3.465 5746580 3.353 3.481 3.353
2512 2015-06-11 3.474 6701280 3.447 3.529 3.436
2511 2015-06-12 3.393 4931020 3.450 3.467 3.385
2510 2015-06-15 3.292 7650700 3.426 3.447 3.286

III) Sorting & Cleaning Data#

Here we remove missing values and add rolling averages (technical indicators).

df.dropna(inplace=True)
df['MA_10'] = df['Close/Last'].rolling(window=10).mean()
df['MA_50'] = df['Close/Last'].rolling(window=50).mean()
df.head()
Date Close/Last Volume Open High Low MA_10 MA_50
2514 2015-06-09 3.344 2886240 3.363 3.379 3.321 NaN NaN
2513 2015-06-10 3.465 5746580 3.353 3.481 3.353 NaN NaN
2512 2015-06-11 3.474 6701280 3.447 3.529 3.436 NaN NaN
2511 2015-06-12 3.393 4931020 3.450 3.467 3.385 NaN NaN
2510 2015-06-15 3.292 7650700 3.426 3.447 3.286 NaN NaN

IV) Data Exploration#

We examine general statistics, visualize stock trends, and analyze feature correlations.

df.describe()
Date Close/Last Volume Open High Low MA_10 MA_50
count 2515 2515.000000 2.515000e+03 2515.000000 2515.000000 2515.000000 2506.000000 2466.000000
mean 2020-06-04 21:46:35.546719488 12.246801 1.684573e+07 12.247913 12.665328 11.837546 12.209585 12.077759
min 2015-06-09 00:00:00 1.165000 3.038000e+04 1.155000 1.216000 0.850000 1.223300 1.383000
25% 2017-12-04 12:00:00 2.304500 2.327225e+06 2.301500 2.346500 2.269500 2.317613 2.313607
50% 2020-06-05 00:00:00 2.894000 4.028380e+06 2.885000 2.950000 2.835000 2.869550 2.777750
75% 2022-12-01 12:00:00 8.051500 1.252550e+07 8.040000 8.224800 7.859000 7.970850 7.719190
max 2025-06-06 00:00:00 118.807000 3.697348e+08 121.200000 122.900000 112.234000 112.198700 94.969060
std NaN 20.981443 3.332872e+07 21.037977 21.822282 20.171035 20.883752 20.516064 
plt.figure(figsize=(12,6))
plt.plot(df['Date'], df['Close/Last'], label='Close')
plt.plot(df['Date'], df['MA_10'], label='10-day MA')
plt.plot(df['Date'], df['MA_50'], label='50-day MA')
plt.title('SMCI Stock Closing Price & Moving Averages')
plt.xlabel('Date')
plt.ylabel('Price')
plt.legend()
plt.grid(True)
plt.show()
../_images/f6f663b3a8ca517a6836b3a7f9c201b32b3bfb5878c3f88408a4b05fdf08e4b1.png 
plt.figure(figsize=(10,5))
sns.histplot(df['Close/Last'], bins=100, kde=True, color='steelblue')
plt.title('Distribution of SMCI Closing Prices')
plt.xlabel('Price (USD)')
plt.ylabel('Frequency')
plt.grid(True)
plt.show() # Histogram

/opt/anaconda3/lib/python3.11/site-packages/seaborn/_oldcore.py:1119: FutureWarning: use_inf_as_na option is deprecated and will be removed in a future version. Convert inf values to NaN before operating instead.
  with pd.option_context('mode.use_inf_as_na', True):
../_images/87f486f3e444b778321671cff11dc2508f90c567cba0dd540287da1d7343b97d.png

Histogram Analysis:
The SMCI closing prices are right-skewed, showing that the stock traded at relatively lower prices most of the time but experienced significant upward spikes. This suggests a volatile growth pattern and a few high-price outliers, especially in recent years.

plt.figure(figsize=(8,6)) # Correlation heatmap
sns.heatmap(df[['Open', 'High', 'Low', 'Close/Last', 'Volume']].corr(), annot=True, cmap='coolwarm')
plt.title('Feature Correlation Heatmap')
plt.show()
../_images/37e3a1d35f4289101bfaba6c6908f457ab2d12d2b79c24e93e42b2e9603870de.png

Heatmap Analysis:
There is strong positive correlation between Open, High, Low, and Close prices, which is expected in stock data. Volume, however, has weaker correlation, suggesting it may have less predictive power for price modeling.

V) Lagged Prices and Moving Averages#

We engineer lagged features and moving averages to help the model learn from previous trends.

for lag in [1, 2, 3]: # Lagged features
    df[f'Lag_{lag}'] = df['Close/Last'].shift(lag)

for ma in [5, 10, 20]: # Lagged features
    df[f'MA_{ma}'] = df['Close/Last'].rolling(window=ma).mean()

df.dropna(inplace=True)
df.head()
Date Close/Last Volume Open High Low MA_10 MA_50 Lag_1 Lag_2 Lag_3 MA_5 MA_20
2465 2015-08-18 2.750 3332050 2.750 2.771 2.7450 2.7170 2.84324 2.765 2.755 2.728 2.7404 2.64695
2464 2015-08-19 2.744 3694050 2.738 2.765 2.6930 2.7123 2.83124 2.750 2.765 2.755 2.7484 2.65720
2463 2015-08-20 2.674 5048290 2.736 2.756 2.6660 2.7110 2.81542 2.744 2.750 2.765 2.7376 2.66210
2462 2015-08-21 2.539 6602320 2.637 2.683 2.5302 2.6991 2.79672 2.674 2.744 2.750 2.6944 2.66180
2461 2015-08-24 2.456 6735940 2.450 2.605 2.3440 2.6729 2.77798 2.539 2.674 2.744 2.6326 2.65860

VI) Linear Regression#

We fit a linear model using key numerical features to predict the next-day price.

df['Next_Close'] = df['Close/Last'].shift(-1) # Features and target
df.dropna(inplace=True)
features = ['Open', 'High', 'Low', 'Close/Last', 'Volume', 'Lag_1', 'MA_5', 'MA_10']
X = df[features]
y = df['Next_Close']

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

model_lr = LinearRegression()
model_lr.fit(X_train, y_train)
y_pred_lr = model_lr.predict(X_test)

print("Linear Regression R²:", r2_score(y_test, y_pred_lr))
print("MSE:", mean_squared_error(y_test, y_pred_lr))

Linear Regression R²: 0.9946140397348965
MSE: 2.6390893816913406

VII) Tree-Based Model (Random Forest)#

To capture nonlinear interactions, we apply a Random Forest Regressor.

model_rf = RandomForestRegressor(n_estimators=100, random_state=42)
model_rf.fit(X_train, y_train)
y_pred_rf = model_rf.predict(X_test)

print("Random Forest R²:", r2_score(y_test, y_pred_rf))
print("MSE:", mean_squared_error(y_test, y_pred_rf))

Random Forest R²: 0.9939059099020344
MSE: 2.986068904520977

  • R² = 0.9939

This means the model explains over 99% of the variance in the next-day SMCI closing price. An R² this high suggests an extremely strong fit, indicating that the model has captured the relationship between input features (price, volume, lags, MAs) and the target variable very well.

  • MSE ≈ 2.99

This is a low value in the context of stock prices (especially if average SMCI prices are much higher). It shows that on average, the predicted price is off by less than $3, which is relatively minor for a volatile tech stock.

Random Forest performed significantly better than the linear regression model, highlighting its strength in handling nonlinear patterns and interactions between features. By incorporating lagged prices and moving averages, the model was able to learn from historical momentum and trends, boosting its predictive power. This result demonstrates the importance of: Feature engineering in time-series problems Using ensemble models for complex, noisy financial data

VIII) Discussion of Results#

  • The linear regression model provides a simple baseline, but underperforms on volatile days.

  • The Random Forest model performs better, likely due to capturing nonlinear price dynamics and lagged features.

  • Lag features and moving averages improve prediction quality.

  • While a high R² is good, it might also signal potential overfitting. This means the model fits the training data too closely and may not generalize as well to unseen market conditions.

  • Financial markets are influenced by external macroeconomic events (news, earnings reports, etc.) not captured in this dataset.

IX) Cross-Validation and Bias–Variance Tradeoff#

In future work, K-Fold Cross Validation can help evaluate model stability.

  • Linear models may underfit due to high bias.

  • Tree-based models may overfit unless tuned, but tend to have lower bias.

X) Summary & Future Improvements#

This project successfully applied EDA and regression to SMCI stock data. Future improvements may include:

  • Cross-validation for robustness

  • Hyperparameter tuning

  • More advanced models (XGBoost, LSTM)

  • Using macroeconomic indicators

XI) References#