Nicotine and caffeine are the most consumed psychostimulant alkaloids for their cognitive enhancement attributes (Burdan, 2015; Harris et al., 2015). These alkaloids have profound effects on the brain and cardiac system and have been well analyzed using bioelectrical signals.
       
The changes in  bioelectrical current and potential give direct and real-time information about body homeostasis and over the years several bioelectrical signals arising from neurons, cardiomyocytes and muscles, the EEG (Electroencephalogram), ECG (Electrocardiogram), EMG (Electromyogram) respectively have evolved into gold standard analytical, diagnostic and prognostic  tools in investigating several physical and mental abnormalities (Pourmohammadi and Maleki, 2020).
       
In the field of biomedical research, the bioelectrical data can be subjected to statistical analysis or machine learning. The statistical analysis is meant for testing hypotheses, generalizing about population based on sample behaviour, drawing conclusions, defining and formalizing threshold values, making inferences through statistical tests (Hazra, 2023). Machine learning algorithms are making great strides in the medical field in recent years (Nguyen et al., 2020). The tools like support vector machine (SVM) are meant for automated binary classification, prediction and forecasting about unobserved outcomes and evaluating best course of action.
       
The Nicotine by acting as agonistic stimulant of pentameric acetylcholine receptors (Mishra et al., 2015) and Caffeine as competitive antagonist of adenosine receptors (Reddy et al., 2024) causes various neurotransmitter and ionic imbalance in central nervous system, peripheral nervous system and cardiac system which have been well analyzed using EEG and ECG (Gilbert et al., 2000; Siepmann et al., 2002; D’Alessandro et al., 2012). But these clinical studies are mostly based on analyses of few features and largely involves use of coffee and cigarettes rather than pure caffeine and nicotine. The brain and cardiac system are involved in constant bidirectional communications to the extent that ECG and EEG signals have shown fusion, coherence and synchronization (Liao et al., 2024). The use of comprehensive sets of several quantitative features of EEG and ECG becomes important to thoroughly document effects of nicotine and caffeine on brain, heart and their bidirectional relationships. The diagnostic methodology among inferential statistics and machine learning algorithms like SVM also needs to be compared. The parametric t test-based inferential statistics for hypothesis testing is meant to draw conclusion, understand significance of the results, formalize and generalize results of sample for population and explore biological explanations. SVM-based supervised machine learning algorithm offers greater flexibility, scalability and accuracy in making predictions and classification using simple mathematical, computational and statistical modelling offering possibility of multimodal signal acquisitions and artificial intelligence based automated diagnostic system designing (Alkhanifer and AlZubi, 2026).
The current piece of research tends to fill this research gap with the objective of:
1. Documenting comprehensive sets of easily extractable quantitative EEG and ECG parameters under effects of chronic nicotine and caffeine in adult male Wistar rats.
2. Comparing inferential statistics and SVM for identifying bioelectrical biomarkers, driving conclusions, making predictions and evaluating diagnostic potential of comprehensive sets of EEG and ECG parameters in nicotine and caffeine administered adult male Wistar rats.
3. Suggesting possibility of dynamic multimodal signal acquisitions, feature selection and artificial intelligence based automated diagnostic system designing.
       
Here, EEG amplitude and frequency based FFT (Fast Fourier transform) parameters were used and conventional amplitude and duration features of ECG were utilized.  For evaluation of their discriminating ability, firstly inferential statistical analysis was conducted using two samples, unpaired t test hypothesizing no difference between means and variance of the experimental groups vis-a-vis control.  Furthermore, the bivariate classification and prediction studies were performed using SVM, a computerized, automated, off-the-shelf, kernel based non-probabilistic, discriminative, supervised machine learning algorithm. 

Animals
 
Thirty adult male Wistar rats, weighing 225–265 g (approx. 14-16 weeks of age), were obtained from the university vivarium of Birla Institute of Technology (BIT, Mesra, Ranchi, India). The rats were randomly divided into three categories (n=10): category I (Control), category II (Nicotine) and category III (Caffeine). All laboratory-based experiments were conducted at the Birla Institute of Technology, Mesra, Ranchi, over a period of two years (July 2022 to July 2024) and data analysis and machine learning experiments were performed in 2025.
 
Drug administration model
 
Laboratory grade 95-99% pure nicotine and caffeine powders were obtained and used for chronic drug administration. The analytical grade nicotine was obtained from Forbes Pharmaceuticals, India and administered at a dose of 0.5 mg/kg body weight (BW) (Xia et al., 2019). The caffeine was brought from Hebansh Pharma Equipment, India and administered at a dose of 10 mg/kg body weight (BW) intraperitoneally (Marin et al., 2011). The rats were randomly divided into 3 categories and chronic administration was done for 21 days. The drug formulations were prepared daily in distilled water, vortexed, warmed and then injected in 0.5 ml quantity for 21 days at the same time every day.  Category I, the Control, was injected with 0.5 M saline, Category II received nicotine intraperitoneally (IP) and Category III was injected with caffeine intraperitoneally.
 
EEG recording
 
The final epidural, single-channel, bipolar EEG recording was performed on the 21^st day under urethane anaesthesia at a dose of 1.2 mg/kg body weight. And the procedure followed was in accordance with previous works (Sinha, 2009). The 2 hours of continuous single channel, bipolar EEG recording was done using Biopac student lab MP45 hardware (Biopac Inc, USA) and the waveforms was obtained in BSL software keeping frequency range of 0.5 to 35 Hz and sampling rate of 256 Hz.
 
EEG preprocessing and analysis
 
EEG preprocessing was done using BSL Software. To remove any kind of artifacts like powerline noise or cardiac activity IIR high + low band-pass digital filter applied at fixed low frequency of 0.5 Hz and high frequency of 40 Hz, thus allowing signal with 0.5 to 40 Hz frequency range to pass. After skipping the first 10 minutes of recording, the entire waveforms of every subject were split into 60 samples each of 2 second epochs.
       
In the analysis, a Hamming window, one of the most widely used windowing methods, was applied, as it provides optimal stop-band attenuation for most digital filter implementations. Linear Fast Fourier transform (FFT) was done (Maheshwari, 2020). The FFT was meant to convert the time domain signal to frequency domain. FFT is an adequate computational algorithm for the estimation of discrete Fourier transform and provides approximate estimation of finite Fourier integrals (Al-Fahoum, 2014).
       
In FFT tab, frequency vs amplitude graph was obtained and four features: maximum frequency, maximum amplitude, average amplitude and area under curve for each considered frequency bands (delta, theta, alpha, beta), were manually obtained and tabulated in Microsoft office excel. Thus, a total of 16 features were taken into consideration.  The choice of considered features was because of their ease of extractions, quantifications, graphical representations and default availability in Biopac software.
 
ECG recording
 
The ECG recording was done simultaneously alongside EEG under deep urethane anaesthesia on the last day of drug exposure past 30 minutes of drug administration. The lead II, surface ECG recording following Einthoven triangle rule, was done using a limb electrode made in the research laboratory by repurposing simple stainless steel paper clips. The recording was done in supine conditions. The signal acquisition was done using a Biopac MP45 hardware system with 0.5 to 35 Hz frequency range and 256 samples/sec sampling rates. The signal was then obtained and processed using BSL software where any movement or other artefacts of the signal were first removed by applying 2 Hz high pass filters. The methodology of electrode designing, data acquisition and preprocessing has been reported previously (Pandey et al., 2025). The amplitudes of the P, R and T deflections, along with the durations of the PR, QRS, QT and RR intervals and heart rate in BPM, were documented, as they are convenient to visualize and quantify. 
 
Statistical analysis and graphical representation
 
All statistical analysis and graphical representations were done using GraphPad Prism version 8.0 software (GraphPad Prism Inc., San Diego, California, USA). All data were presented as mean with standard error of mean (SEM).  A two tailed unpaired or independent t test using a set of continuous variables, i.e., considered EEG and ECG parameters, was done for each set of experiments-Experiment group I: Control vs Nicotine and Experiment group II: Control vs Caffeine. A p value of 0.05 or less was considered statistically significant.

Inferential statistic
 
In our experiment, inferential statistics is important as the set of EEG, ECG parameters used to compare between control and experimental groups have been used comprehensively for the first time. So, quantitative estimation of difference between control and psychostimulant groups can help make generalized inferences and explore the causes. The bivariate comparison between control and stimulant administered groups was done using parametric, two sample, unpaired student t-test as samples are independent, symmetrical and have equal variance. The difference between two groups was considered significant when the probability of similarity between the values of means is less than 0.05 or 5% (Schober and Vetter, 2019).
 
Support vector machine
 
The support vector machine (SVM) has shown a high degree of utility in solving classification problems for several neuropsychic pathophysiology ailments using EEG and ECG parameters (Ancillon et al., 2022). In present work, we have used simple, straightforward linear kernel based SVM to identify the best hyperplane with the largest margin between two classes for binary classification in two sets of experiments separately, Group I Experiment: control and chronic Nicotine and Group II Experiment: control and chronic Caffeine using EEG and ECG variables. Here, we collected 10 data instances from each rat, a total of 100 from each group in each set of experiments and used 80% of datasets for training the SVM algorithm and 20% for its validation. SVM maps an input vector X into a high dimensional feature space Z through a mapping function theta. The hyperplane separates data belonging to two different classes: Control or caffeine and control or nicotine in Z.
       
The performance of the SVM classifier was tested using a confusion matrix which calculated accuracy, precision, recall and F1-score. These matrices were used to provide elaborate and relevant assessment of the model’s predictive ability to detect chronic drug exposure. Here, accuracy gives a general performance overview of the model, precision measures the number of true predictions redeemed upon total retrieved instances. Recall measures the number of correct instances retrieved divided by total correct instances, F1 score balances the precision and recall and is calculated as a harmonic mean of precision and recall (Valero-Carreras et al., 2023).

Inferential statistics
 
For comparing control and stimulant groups, the FFT based features of EEG in 4 frequency bands and morphological features, i.e., amplitude and duration of different peaks in ECG bands, were considered as these features have well optimized and familiar methods of extraction, calculations and interpretations.
 
A. Group I Experiment: Control Vs Nicotine EEG data
 
In the Control vs Nicotine group, in the EEG data  (Table 1), in the delta frequency band (Fig 1: A1 to A4), the maximum frequency significantly decreased (T = 6.06) and p being <0.0001, the increase in maximum and mean amplitude, area under curve were also significant (p<0.01). In the theta frequency band (Fig 1: B1 to B4),  only the maximum frequency parameter showed significant increase (T = 5.362, p<0.0001) whereas amplitude and area parameters showed insignificant departure. In the alpha wave band (Fig 1: C1 to C4), all 4 parameters showed significant decrease with p being <0.0001. In the Beta wave frequency band (Fig 1: D1 to D4), all 4 parameters showed significant departure in nicotine vis-α-vis control (p<0.0001).




 
B. Group I Experiment: Control Vs Nicotine ECG data
 
In Control vs Nicotine ECG morphology (Table 2) (Fig 2: A1 to A8), there were significant increase in amplitude of P wave (T= 6.549) (Fig 2: A1), T wave (T = 8.892) (Fig 2: A3), heart rate (Fig 2:A4) and significant decrease in RR interval (T= 12.7) (Fig 2:A8). Whereas only marginal variations were seen in amplitude of R wave and PR, QRS and QT intervals. 




 
C. Group 2 Experiment: Control Vs Caffeine EEG data
 
In caffeine category the EEG (Table 3, Fig 3), delta wave band (Fig 3: A1 to A4) showed significant decrease in dominant frequency (T=7.67, p<0.0001) (Fig 3: A1) and significant increase in maximum amplitude (T = 16.4), mean amplitude (T = 4.129) and area (T = 11.69) with p value less than 0.0001 in all three cases. In Theta frequency band (Fig 3: B1 to B4), two amplitude parameters out of 4 parameters showed significant increase in caffeine when compared to control groups. While P remained <0.0001, T value for maximum and mean amplitude were found to be 3.088 and 4.129 respectively. In the Alpha wave band (Fig 3: C1 to C4), the dominant frequency (T= 3) and area (T = 5.191) decreased significantly p<0.0001 in the caffeine group whereas only marginal decreases in maximum and mean amplitude were seen. In the beta wave band (Fig 3: D1 to D4), the decrease in maximum amplitude (T = 8.45) and area (T = 9.677) were significant with P being less than 0.0001 in both the cases.




 
D. Experiment group II: Control vs Caffeine ECG data
 
In Control vs caffeine group, the morphological parameters of ECG (Table 4) (Fig 4 A1 to A8) that showed significant (p<0.0001) increase, included P wave amplitude (T = 5.462), T (T = 5.507), Heart rate (T = 13.89), PR interval (T = 10.02), QRS interval (T =4.168), QT interval (T = 7.035) and decreased RR interval (T = 12.61).





Support vector machine results
 
The SVM has deep-rooted foundations in mathematics, statistics, optimization and machine learning. Its performance in automated binary classification of nicotine and caffeine groups for both EEG and ECG was evaluated using Accuracy, Precision, Recall and F1 score.
 
A. Group I Experiment: Control vs Nicotine group
 
SVM results for “Control vs Nicotine group” are shown in Table 5. The EEG parameters showed accuracy of 99.63%, precision calculated to be 99.4%, recall was 99.46 and F1 score was found to be 99.63%. The ECG parameters have test accuracy of 96.087%, precision was 94.05%, recall found to be 99.35% and F1 score calculated to be 96.625%.


 
B. Group 2 Experiment: Control vs Caffeine group
 
SVM results for  “Control vs Caffeine group” are shown in Table 6. The EEG parameters showed test accuracy of 99.45%, precision being 99.475%, recall was found to be 99.46% and F1 score was obtained to be 99.4697%. In case of ECG parameters, the test accuracy was 96.07%, precision being 96.26%, recall was 95.49% and F1 score was found to be 95.856%.


               
The result hails that the popular natural psychostimulants nicotine and caffeine have profound influence on brain and heart bioelectrical signals, the EEG and ECG respectively and use of student t test and SVM can help in hypothesis testing, diagnosis, predictions and classification. The research made use of bioelectrical signals, the EEG and ECG, for investigation. These signals are easy to acquire and deliver robust and powerful information. These have stood firm in the test of time garnering modern scientific validation in age of big data and automated sensors. Finally, the research also tends to compare and evaluate the student t test-based inferential statistical hypothesis testing and SVM-based supervised machine learning algorithm in pinpointing potent biomarkers that can be used to solve problems related to classification, prediction and scientific interpretations.

The research concludes by documenting that Nicotine and Caffeine have profound effects on easily extractable quantitative EEG and ECG parameters and advocates for research and development of single wearables for dynamic multimodal bio-electrophysiological monitoring in precise and coherent manner for making inference and automated classification. The research analyses the importance of t test and SVM. Therefore, research opines for simultaneous and complementary use of both the methods to reach biologically interpretable, generalizable outcomes that can be used in embedded wearables for automated classification and predications. The research also suggests comprehensive molecular investigations to understand the reason behind certain observations which shows stark departure from erstwhile research.

The authors thank Dr. Ashok Patnaik, Associate Professor, Department of Pharmaceutical Science and Technology, BIT, Mesra, Ranchi for timely ethical clearance of the research.
 
Disclaimers
 
The views and conclusions expressed in this article are solely those of the authors and do not necessarily represent the views of their affiliated institutions. The authors are responsible for the accuracy and completeness of the information provided, but do not accept any liability for any direct or indirect losses resulting from the use of this content.
 
Informed consent (ethical approval)
 
The study complies to standard national and international ethical norms after getting due approval from institute ethical approval committees (IEAC) of Birla Institute of Technology (BIT), Mesra, Ranchi, India (Approval number: 1972/PH/BIT/03/23/IAEC).

The authors declare that there are no conflicts of interest regarding the publication of this article. No funding or sponsorship influenced the design of the study, data collection, analysis, decision to publish, or preparation of the manuscript.